bsc-pm / bots Goto Github PK

There are a several calls to alloca and malloc that try to store the results to non void * variables without casting appropriately. For example in serial/nqueens/nqueens.c line 106, in the find_queens function, the code does this:

a = alloca(size * sizeof(char));

Where a is a char *. The Intel compiler does not like this.

Strassen should compute a matrix multiplication C = A * B. Right now, matrices A and B are initialized, but then both strassen_main_par and strassen_main_seq are called to compute A = B * C instead. And since C is uninitialized, the result will mostly be 0.

If using validation, compare_matrix will calculate the relative difference between two cells of the parallel and sequential result. For most cells this will be (0-0)/0, which is of course undefined and passes the validation for whatever reason.

A simple fix could be to call the main functions as strassen_main_par(A,B,C,bots_arg_size) instead of strassen_main_par(C, A, B, bots_arg_size) in app-desc.h.

Hi, I have been running the bots applications, including Health, for a paper I am writing. I have noticed that if Health is run with the -check flag, the verification is always unsuccessful. I have tried using -check on fft and sort, and there it is successful.

How to reproduce:

1. Git pull latest version (master) or download older bots-1.1.2.tar.gz and extract
2. ./configure and make on a modern 64-bit setup. I have tried using both gcc and Clang.
3. cd run && ./run-health.sh -check

After going through the source, and re-checking compiler output, this is the culprit:
https://github.com/bsc-pm/bots/blob/master/omp-tasks/health/health.c#L561
Which gcc does complain about:

health.c:564:21: warning: format ‘%ld’ expects argument of type ‘long int *’, but argument 9 has type ‘int32_t * {aka int *}’ [-Wformat=]

For medium inputs, the parameters are incorrectly parsed as follows:

Number of levels    = 4
Cities per level    = 36
Population ratio    = 10
Simulation time     = 0    <-- Should be 365
Assess time         = 2
Convalescence time  = 12
Initial seed        = 23
Get sick prob.      = 0.002000
Convalescence prob. = 0.100000
Realloc prob.       = 0.150000

Changing the %ld format to %d fixes the problem. I assume it was %ld because it is an old application, and because the seed needed to be atleast 32 bits in the days where it was only guaranteed to be 16 bits. However, on modern setups, this causes a bug.

Hi,
apparently you forgot to add bin/* to the gitignore file. Could you please do that?

Thanks.

Health and nQueens sources leads to produce always executables with UNTIED tasks when compiling with IF-Clause or Final Cut-Off, even when executables with TIED tasks should be produced.

I generated some dump of the object files and indeed the "flag" parameter passed to function "GOMP_task" has always the UNTIED bit set in the aforementioned scenarios.