Enabling RNG based on oneMKL generate_single

[L30nardoSV]

About

This PR proposes using oneMKL RNG generators as alternatives RNGs in device code. So far only the generate_single() is employed.

• By default, the original RNGs are used.
• To select an oneMKL RNG engine, uncomment the corresponding line in dpcpp/kernels.dp.cpp:

//#define RNG_ONEMKL_MRG32K3A
//#define RNG_ONEMKL_PHILOX4X32X10
//#define RNG_ONEMKL_MCG31M1
//#define RNG_ONEMKL_MCG59

Evaluation on DevCloud Gen9 GPU: docking time (s)

Early-termination enabled

make DEVICE=XeGPU TEST=sw/ad test

	SW	AD
Original	58.99	45.98
MRG32K3A	65.51	57.89
PHILOX4X32X10	60.58	48.15
MCG31M1	60.09	46.62
MCG59	56.63	47.91

Early-termination disabled

• make DEVICE=XeGPU
• ./bin/autodock_xegpu_64wi -ffile ./input/3ce3/derived/3ce3_protein.maps.fld -lfile ./input/3ce3/derived/3ce3_ligand.pdbqt -nrun 20 -ngen 27000 -psize 150 -resnam test -gfpop 0 -A 0 -H 0 -lsmet sw/ad

	SW	AD
Original	43.07	95.97
MRG32K3A	62.39	112.61
PHILOX4X32X10	44.21	96.93
MCG31M1	43.29	96.74
MCG59	43.38	96.59