CUDA version lulesh.cu CalcTimeConstraintsForElems function, stream usage

Nobody has claimed this yet.

Vector_d* dev_mindtcourant= Allocator< Vector_d >::allocate(dimGrid);
Vector_d* dev_mindthydro = Allocator< Vector_d >::allocate(dimGrid);

CalcTimeConstraintsForElems_kernel <<>>
(length,qqc2,dvovmax,
domain->matElemlist.raw(),domain->ss.raw(),domain->vdov.raw(),domain->arealg.raw(),
dev_mindtcourant->raw(),dev_mindthydro->raw());

// TODO: if dimGrid < 1024, should launch less threads
CalcMinDtOneBlock <<<2,max_dimGrid, max_dimGrid*sizeof(Real_t), domain->streams[1]>>>(dev_mindthydro->raw(),dev_mindtcourant->raw(),domain->dtcourant_h,domain->dthydro_h, dimGrid);

cudaEventRecord(domain->time_constraint_computed,domain->streams[1]);

Allocator >::free(dev_mindtcourant,dimGrid);
Allocator >::free(dev_mindthydro,dimGrid);

I am not sure, but wouldn't it be an issue as dev_mindtcourant is used in streams[1] and will be free in the default stream?

Start in lulesh.cu at CalcTimeConstraintsForElems and trace the allocation, kernel launch, event recording, and frees for dev_mindtcourant and dev_mindthydro. Check the CUDA stream behavior around domain->streams[1] and determine whether the reported free is ordered safely; done means the lifetime is correct or the issue is documented with evidence.

Written by the indexing model from the issue text.