Re: [Starpu-devel] Issues met during installing and configuration of StarPU

[Samuel Thibault <samuel.thibault@inria.fr>]

Hello,

11610515, le mar. 12 mars 2019 17:32:23 +0800, a ecrit:
>         1.We can compile and run most of the code under "basic_example"
> following the document with command:
>            $ gcc mult.c `pkg-config starpu-1.2 --cflags` `pkg-config 
> starpu-1.2
> --libs` -lm

Ok, now I understand the issue.  You do not need to compile examples by
hand, you can just run make inside the examples/ directory and they will
get compiled.

>          With warnings in runtime:
>          [starpu][initialize_eager_center_policy] Warning: you are running 
> the
> default eager scheduler, which is not a very smart scheduler. Make sure to 
> read
> the StarPU documentation about adding performance models in order to be 
> able to
> use the dmda or dmdas scheduler instead.
>           Which we think is normal since we haven't done anything to the
> scheduler.

Indeed.

>         2.We met compile errors when compiling some of the examples under
> "basic_example" with
>            $ gcc vector_scal.c `pkg-config starpu-1.2 --cflags` `pkg-config
> starpu-1.2 --libs` -lm
>               /tmp/ccTzBv7f.o:(.data+0xf0): undefined reference to
> `scal_cpu_func'

That is not surprising: that example is composed of multiple .c files
that you need to link altogether. Again, just running make will get it
linked appropriately.

>              But others executed with problems
>              $ ./hello_world_top
>                [starpu][initialize_eager_center_policy] Warning: you are
> running the default eager scheduler, which is not a very smart scheduler. 
> Make
> sure to read the StarPU documentation about                         adding
> performance models in order to be able to use the dmda or dmdas scheduler
> instead.
>                 ../../src/top/starpu_top.c:160 launching network threads
>                 ../../src/top/starpu_top_connection.c:120 Connection to UI
> initialization

That example requires the launch of the starpu-top alongside.

>               $ ./pi     
>                  [starpu][initialize_eager_center_policy] Warning: you are
> running the default eager scheduler, which is not a very smart scheduler. 
> Make
> sure to read the StarPU documentation about                         adding
> performance models in order to be able to use the dmda or dmdas scheduler
> instead.
>                  [1]    34191 segmentation fault (core dumped)  ./pi

That, however, is surprising, it just works here.  It would be useful to
take a backtrace of the crash.

>               $ ./yuv_downscaler
>                   yuv_downscaler: ../../examples/ppm_downscaler/
> yuv_downscaler.c:129: main: Assertion `(filesize % sizeof(struct 
> yuv_frame)) ==
> 0' failed.
>                  [1]    34672 abort (core dumped)  ./yuv_downscaler

That example needs data, which is not included in the tarball indeed
(and my student didn't take the time to write at least some hints about
that, now done).

>              $ ./tasks_size_overhead.gp
>                ./tasks_size_overhead.gp: 21: ./tasks_size_overhead.gp: 
> cannot
> open tasks_size_overhead.output: No such file

That is just a script that gets run by ./tasks_size_overhead.sh, which
you should use instead.

Samuel