openmp & osx M2 : ok with the script

[alaingdl]

I am happy to say that I can see OpenMP is activated at compilation on M2 with OSX with a GDL compile only with the script, without any change. And it sensitive to the number of cores we request. Great !

I am not sure to fully understood some numbers but it will be another story !

GDL> BASIC_BENCHMARKS,nbp=1e9
% Compiled module: BASIC_BENCHMARKS.
Initial: nb detected CORES :    8, nb cores/threads in use :    8
Working on type : FLOAT
Time for RANDOMN (1/4)      3.21649
Time for RANDOMU :          3.10277
Time for COS :             0.644208
Time for SIN :             0.654338
Time for TAN :             0.785514
Time for ALOG :            0.523582
Time for SQRT :            0.196355
Time for EXP :             0.410711
Time for WHERE :           0.982324
Time for SORT (1/30) :      2.88457
Time for FINITE :          0.177522
Time for FIX :             0.118620
% Compiled module: BENCHMARK_GENERATE_FILENAME.
% Compiled module: GDL_IDL_FL.
% Compiled module: BENCHMARK_INFO_CPU.
% Compiled module: BENCHMARK_INFO_OS.
% Compiled module: BENCHMARK_INFO_SOFT.
Writing file : bench_basic_GDL_osxm2mutu.obspm.fr_y2024m10d24_20:33:12.xdr
GDL> BASIC_BENCHMARKS,nbp=1e9, set=1
Initial: nb detected CORES :    8, nb cores/threads in use :    8
Set: nb detected CORES :    8, nb cores/threads in use :    1
Working on type : FLOAT
Time for RANDOMN (1/4)      3.24336
Time for RANDOMU :          3.10482
Time for COS :              3.44342
Time for SIN :              3.33776
Time for TAN :              3.88861
Time for ALOG :             2.25250
Time for SQRT :            0.397583
Time for EXP :              1.71939
Time for WHERE :           0.477414
Time for SORT (1/30) :      4.98597
Time for FINITE :          0.601096
Time for FIX :             0.202139
Writing file : bench_basic_GDL_osxm2mutu.obspm.fr_y2024m10d24_20:36:58.xdr
GDL> 

[GillesDuvert]

RANDOM[N|U] does not use threads (unless you replace src/randomgenerators.cpp by the content of resource/randomgenerators_parallel.cpp, but the (huge) speedup is only reached after initialisation of parallel 'seeds', which is not what benchmarks test.
All functions that are not a simple math lib call ( WHERE, as opposed to COS) have a minimum time imposed by the creation/destruction of GDL variables, not by the parallel processing.

[pjb7687]

Just now, I've uploaded binaries (Windows and macOS) to the latest v1.1 release.