Improve helicity filtering

[valassi]

In studying issue #16, I just realised that the 274 requests to access global memory from each warp reduce to 16 (!) if I disable helicity filtering. There is no other indication that this is an issue (andactually the present implementation saves a factor 2 in CUDA and a factor 4 in CPP), but it's worth having a look...

[valassi]

Ok this is done - though ugly and can be improved, and certainly code should be repackaged: 194ffc1

Ugly but effective.

• It gets back the speedup of 2 on GPU (keeps the old implementation on CPP). So throughput is again in the 6E8+.
• Also, it reduces the number of requests for global memory. This may not be a performance improvement but makes the analysis much simpler.

time ./gcheck.exe -p 16384 32 12
***************************************
NumIterations             = 12
NumThreadsPerBlock        = 32
NumBlocksPerGrid          = 16384
---------------------------------------
FP precision              = DOUBLE (nan=0)
Complex type              = THRUST::COMPLEX
RanNumb memory layout     = AOSOA[4]
Momenta memory layout     = AOSOA[4]
Wavefunction GPU memory   = LOCAL
Curand generation         = DEVICE (CUDA code)
---------------------------------------
NumberOfEntries           = 12
TotalTimeInWaveFuncs      = 1.039935e-02 sec
MeanTimeInWaveFuncs       = 8.666127e-04 sec
StdDevTimeInWaveFuncs     = 4.685819e-06 sec
MinTimeInWaveFuncs        = 8.582820e-04 sec
MaxTimeInWaveFuncs        = 8.754520e-04 sec
---------------------------------------
TotalEventsComputed       = 6291456
RamboEventsPerSec         = 6.196162e+07 sec^-1
MatrixElemEventsPerSec    = 6.049854e+08 sec^-1
***************************************
NumMatrixElements(notNan) = 6291456
MeanMatrixElemValue       = 1.372152e-02 GeV^0
StdErrMatrixElemValue     = 3.269516e-06 GeV^0
StdDevMatrixElemValue     = 8.200854e-03 GeV^0
MinMatrixElemValue        = 6.071582e-03 GeV^0
MaxMatrixElemValue        = 3.374925e-02 GeV^0
***************************************
00 CudaFree : 0.144857 sec
0a ProcInit : 0.000547 sec
0b MemAlloc : 0.081633 sec
0c GenCreat : 0.015051 sec
1a GenSeed  : 0.000014 sec
1b GenRnGen : 0.007956 sec
2a RamboIni : 0.000133 sec
2b RamboFin : 0.000083 sec
2c CpDTHwgt : 0.008282 sec
2d CpDTHmom : 0.093040 sec
3a SGoodHel : 0.001629 sec
3b SigmaKin : 0.000088 sec
3c CpDTHmes : 0.010311 sec
4a DumpLoop : 0.023784 sec
9a DumpAll  : 0.023715 sec
9b GenDestr : 0.000061 sec
9c MemFree  : 0.025542 sec
9d CudReset : 0.043790 sec
TOTAL       : 0.480517 sec
TOTAL(n-2)  : 0.291870 sec
***************************************
real    0m0.491s
user    0m0.182s
sys     0m0.308s

./profile.sh -nogui -p 1 4 1
  gProc::sigmaKin(double const*, double*), 2020-Aug-18 09:25:37, Context 1, Stream 7
    Section: Command line profiler metrics
    ---------------------------------------------------------------------- --------------- ------------------------------
    l1tex__t_requests_pipe_lsu_mem_global_op_ld.sum                                request                             16
    l1tex__t_sectors_pipe_lsu_mem_global_op_ld.sum                                  sector                             16
    ---------------------------------------------------------------------- --------------- ------------------------------