- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Dear all,
I have an issue compiling and running the Fortran 90 version of the EPCC Microbenchmarks, which can be downloaded here:
http://www.epcc.ed.ac.uk/computing/research_activities/openmpbench/download.html
I can correctly compile the 'arraybench' benchmark, but when executing it, the program aborts while testing the FIRSTPRIVATE clause. I use the -openmp switch to the Fortran compiler (both version 8.1.024 and the evaluation version 9.0.021) and the problem shows up independently of all other options I use (-g, -O2).
To be more precise, the above benchmark can be compiled for various sizes of arrays. The Makefile contains all powers of 3, from 1 up to 59049. If the size of the array is chosen from those predefined values and is less or equal to 243(= 3^5), the benchmark completes successfully. If the size is greater than 243, the benchmark completes successfully only if it is run with one thread. If more threads are used, it crashes while testing the FIRSTPRIVATE clause.
I tried to find the biggest size of the array that runs correctly and I found that it was 255, which seems suspicious (= 2^8 - 1).
Firstly, I though that there is not enough stack to run the benchmark with big sizes, but a ulimit -a shows that the stack size is unlimited. I also tried to use the environment variable KMP_STACKSIZE and have set it to the biggest value allowed in my environment (~290MB), but to no avail.
Any ideas what else I could try?
Ioannis
I have an issue compiling and running the Fortran 90 version of the EPCC Microbenchmarks, which can be downloaded here:
http://www.epcc.ed.ac.uk/computing/research_activities/openmpbench/download.html
I can correctly compile the 'arraybench' benchmark, but when executing it, the program aborts while testing the FIRSTPRIVATE clause. I use the -openmp switch to the Fortran compiler (both version 8.1.024 and the evaluation version 9.0.021) and the problem shows up independently of all other options I use (-g, -O2).
To be more precise, the above benchmark can be compiled for various sizes of arrays. The Makefile contains all powers of 3, from 1 up to 59049. If the size of the array is chosen from those predefined values and is less or equal to 243(= 3^5), the benchmark completes successfully. If the size is greater than 243, the benchmark completes successfully only if it is run with one thread. If more threads are used, it crashes while testing the FIRSTPRIVATE clause.
I tried to find the biggest size of the array that runs correctly and I found that it was 255, which seems suspicious (= 2^8 - 1).
Firstly, I though that there is not enough stack to run the benchmark with big sizes, but a ulimit -a shows that the stack size is unlimited. I also tried to use the environment variable KMP_STACKSIZE and have set it to the biggest value allowed in my environment (~290MB), but to no avail.
Any ideas what else I could try?
Ioannis
Link Copied
2 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello,
I submitted this problem to Intel Premier quite some time ago, and at
least for the test case I gave the most recent 8.1 and 9.0 compiler
releases were OK. I haven't tried the complete EPCC since then, though.
Regards
I submitted this problem to Intel Premier quite some time ago, and at
least for the test case I gave the most recent 8.1 and 9.0 compiler
releases were OK. I haven't tried the complete EPCC since then, though.
Regards
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hello,
Intel has announced that they will release the non-commercial version of the compiler at some time in August. Probably they will have fixed some issues until then and therefore I am waiting for this release. If, however, the problem persists with that version, I will try to make the smallest possible test case and submit a problem report to Intel Premier.
Ioannis
Intel has announced that they will release the non-commercial version of the compiler at some time in August. Probably they will have fixed some issues until then and therefore I am waiting for this release. If, however, the problem persists with that version, I will try to make the smallest possible test case and submit a problem report to Intel Premier.
Ioannis
Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page