Compile code to be relocatable. #261
Comments
|
Do you need this through spack-stack? If you just need it for your own development copy of ip, then building with |
|
The goal is to make python call libwgrib2.so (import pywgrib2_s). This
means that libraries
that wgrib2 calls will have to be compiled with -fPIC.
…On Fri, Nov 29, 2024 at 6:07 PM Alex Richert ***@***.***> wrote:
Do you need this through spack-stack? If you just need it for your own
development copy of ip, then building with
-DCMAKE_POSITION_INDEPENDENT_CODE=ON should do the trick (I assume you're
compiling a static library).
—
Reply to this email directly, view it on GitHub
<#261 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AEIB7ZV2W4JUEMQLILZM7MT2DDXR5AVCNFSM6AAAAABSX2MFM6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMBYG4ZDINRSGQ>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
|
Have you already tried |
|
I got all the netcdf, hdf5, g2c, png, jasper, openjpeg, aec libraries to
work.
IP was the only exception. I can make wgrib2 when I turned off
MAKE_SHARED_LIB
in wgrib2.
…On Mon, Dec 2, 2024 at 1:59 PM Alex Richert ***@***.***> wrote:
Have you already tried -DCMAKE_POSITION_INDEPENDENT_CODE=ON for wgrib2?
If that doesn't work then that's probably a cmake bug for wgrib2.
—
Reply to this email directly, view it on GitHub
<#261 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AEIB7ZX4SS4FRHSDGGJ7EPL2DSUYBAVCNFSM6AAAAABSX2MFM6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMJSGQ3DIMJUG4>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
|
To investigate further I'll need exact steps to reproduce the problem, please. |
|
How I tell whether *.a is relocatable:
$ ar x libip_4.a
$ gcc -shared -o libtest.so *.o
Complaints about -fPIC.
I built ip by:
git clone "https://github.com/NOAA-EMC/NCEPLIBS-ip"
cd NCEPLIB-ip
edit CMakeLists.txt
change option(OPENMP "Use OpenMP threading" OFF)
to option(OPENMP "Use OpenMP threading" ON)
mkdir build
cd build
make
sudo make install
…On Mon, Dec 2, 2024 at 2:56 PM Alex Richert ***@***.***> wrote:
To investigate further I'll need exact steps to reproduce the problem,
please.
—
Reply to this email directly, view it on GitHub
<#261 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AEIB7ZUY6JOKWLW6EXOSSN32DS3M5AVCNFSM6AAAAABSX2MFM6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMJSGY2DENBQGQ>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
|
I don't see the cmake step-- are you using |
|
No. So I have to make a non-default version of IP.
…On Mon, Dec 2, 2024 at 3:54 PM Alex Richert ***@***.***> wrote:
I don't see the cmake step-- are you using
-DCMAKE_POSITION_INDEPENDENT_CODE=ON? i.e., cmake ..
-DCMAKE_POSITION_INDEPENDENT_CODE=ON?
—
Reply to this email directly, view it on GitHub
<#261 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AEIB7ZQFLYY6ITPJBK4P6ML2DTCJ5AVCNFSM6AAAAABSX2MFM6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMJSHAYDQNJRGU>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
|
Correct, it'll require either explicit PIC with the above build option, or building the shared version. |
|
Of course it works if you rebuild the IP package
with -DCMAKE_POSITION_INDEPENDENT_CODE=ON.
The point is that the default cmake options does not produce a library that
is usable in some
applications. One could argue that the position independent code is a
unusual feature and
is not worth the performance penalty. Anyways I need to know the IPlib
policy for future wgrib2
documentation.
…On Mon, Dec 2, 2024 at 5:08 PM Alex Richert ***@***.***> wrote:
Correct, it'll require either explicit PIC with the above build option, or
building the shared version.
—
Reply to this email directly, view it on GitHub
<#261 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AEIB7ZXYYBJSZBDQWCRLBZD2DTK35AVCNFSM6AAAAABSX2MFM6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMJTGA2DENZZGA>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
|
I'd be open to changing the default if it doesn't have a significant impact on performance. In the case of, say, g2c, I would have expected that it would have the same issue. Which version of ip are you using, is this the head of develop? |
|
Performance is important but people tend not to worry about ip and g2/g2c
because
those are used in I/O bound processes (at least for the models). The time
sensitive
NCEP uses of wgrib2 are NOMADS and the gfs post-processing. The NOMADS job
doesn't
use IP. The post-processing uses IP but tends to be I/O bound.
G2c is position independent, and I am using the develop version of IP.
Openblas,
PNG, AEC libraries from Ubuntu were position independent. The Jasper build
was by default position independent.
…On Tue, Dec 3, 2024 at 5:11 PM Alex Richert ***@***.***> wrote:
I'd be open to changing the default if it doesn't have a significant
impact on performance. In the case of, say, g2c, I would have expected that
it would have the same issue. Which version of ip are you using, is this
the head of develop?
—
Reply to this email directly, view it on GitHub
<#261 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AEIB7ZWWNVKBEHZ35Z67XAT2DYUCRAVCNFSM6AAAAABSX2MFM6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMJVGY2TKMRVGY>
.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Compile with -fPIC to get relocatable code. Want to make a (relocatable) python library that uses iplib.
The text was updated successfully, but these errors were encountered: