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

["11610515" <11610515@mail.sustech.edu.cn>]

Dear Professor Thibault and Starpu Team:

        Thanks for your reply and kind instruction. And I'm going to describe the problems in detail in the following. The config.log file is included in the attachment.

        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

         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.

        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'
             /tmp/ccTzBv7f.o:(.data+0xf8): undefined reference to `scal_sse_func'
             /tmp/ccTzBv7f.o:(.data+0x110): undefined reference to `scal_cuda_func'
             collect2: error: ld returned 1 exit status

          We used nm command under the lib and found out that these functions were not implemented:

             # undergrats @ sustc_lab_1009 in ~/tools/starpu_cuda/lib [22:11:06]
             $ nm *.so | grep "scal_cpu_func"

             # undergrats @ sustc_lab_1009 in ~/tools/starpu_cuda/lib [22:11:38] C:1
             $ nm *.so | grep "scal_sse_func"

             # undergrats @ sustc_lab_1009 in ~/tools/starpu_cuda/lib [22:11:46] C:1
             $ nm *.so | grep "scal_cuda_func"

       

             3.Most of the pre-compiled binary file including vector_scal can be executed and ran smoothly, so did other example binary files:

              # undergrats @ sustc_lab_1009 in ~/tools/starpu_cuda/lib/starpu/examples [22:13:02]

              $ ls

               (binary files....)

             # undergrats @ sustc_lab_1009 in ~/tools/starpu_cuda/lib/starpu/examples [22:13:03]
             $ ./vector_scal
              [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.
              [BEFORE] 1-th element    : 2.00
              [BEFORE] (NX-1)th element: 204800.00
              [AFTER] 1-th element     : 6.28 (should be 6.28)
              [AFTER] (NX-1)-th element: 643072.00 (should be 643072.00)
              [AFTER] Computation is correct

             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
                ^C  (The program stuck at this stage)

               

              $ ./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

              $ ./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

             

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

               gnuplot> plot  x title "linear",  "tasks_size_overhead.output" using 1:()/($23) with linespoints title columnheader(22),  "tasks_size_overhead.output" using 1:()/($21) with linespoints title                  columnheader(20),  "tasks_size_overhead.output" using 1:()/($19) with linespoints title columnheader(18),  "tasks_size_overhead.output" using 1:()/($17) with linespoints title                                columnheader(16),  "tasks_size_overhead.output" using 1:()/($15) with linespoints title columnheader(14),  "tasks_size_overhead.output" using 1:()/($13) with linespoints title                                columnheader(12),  "tasks_size_overhead.output" using 1:()/($11) with linespoints title columnheader(10),  "tasks_size_overhead.output" using 1:()/($9) with linespoints title                                  columnheader(8),  "tasks_size_overhead.output" using 1:()/($7) with linespoints title columnheader(6),  "tasks_size_overhead.output" using 1:()/($5) with linespoints title                                        columnheader(4),  "tasks_size_overhead.output" using 1:()/($3) with linespoints title columnheader(2)
                                                                        ^
                line 0: invalid _expression_

               I think this one is related to gnuplot and may have less connection with our starpu problem, but for the integrality of the report I put in on.

     We ran these programs on the same device with ubuntu 16.04 LTS, Cuda V9.0.176, 8* Nvidia GTX 1080 Ti GPUs, 4* Intel Xeon E5-2640 v4 CPUs and we did not install OpenCL. The Starpu version is 1.2.8 and was installed from source.

         

Attachment: config.log
Description: Binary data