NBodyStatus nbInitNBodyStateCL(NBodyState* st, const NBodyCtx* ctx, const CLRequest* clr)
{
cl_int err;
/* Bodies must be set before trying to use this */
if (!st->bodytab)
{
return NBODY_CONSISTENCY_ERROR;
}
if (ctx->potentialType == EXTERNAL_POTENTIAL_CUSTOM_LUA)
{
mw_printf("Cannot use Lua potential with OpenCL\n");
return NBODY_UNSUPPORTED;
}
st->usesQuad = ctx->useQuad;
st->usesExact = (ctx->criterion == Exact);
st->usesCL = TRUE;
st->ci = mwCalloc(1, sizeof(CLInfo));
st->nbb = mwCalloc(1, sizeof(NBodyBuffers));
st->workSizes = mwCalloc(1, sizeof(NBodyWorkSizes));
st->kernels = mwCalloc(1, sizeof(NBodyKernels));
err = mwSetupCL(st->ci, clr);
if (err != CL_SUCCESS)
return NBODY_CL_ERROR;
if (!nbCheckDevCapabilities(&st->ci->di, ctx, st->nbody))
return NBODY_CAPABILITY_ERROR;
if (nbSetThreadCounts(st->workSizes, &st->ci->di, ctx) || nbSetWorkSizes(st->workSizes, &st->ci->di))
return NBODY_ERROR;
st->effNBody = nbFindEffectiveNBody(st->workSizes, st->usesExact, st->nbody);
if (nbLoadKernels(ctx, st))
return NBODY_CL_ERROR;
err = nbCreateBuffers(ctx, st);
if (err != CL_SUCCESS)
return NBODY_CL_ERROR;
err = nbSetInitialTreeStatus(st);
if (err != CL_SUCCESS)
return NBODY_CL_ERROR;
err = nbSetAllKernelArguments(st);
if (err != CL_SUCCESS)
return NBODY_CL_ERROR;
err = nbMarshalBodies(st, CL_TRUE);
if (err != CL_SUCCESS)
return NBODY_CL_ERROR;
return NBODY_SUCCESS;
}
NBodyStatus nbInitNBodyStateCL(NBodyState* st, const NBodyCtx* ctx)
{
cl_int err;
const DevInfo* devInfo;
if (!st->usesCL)
{
mw_printf("CL not setup for CL state initialization\n");
return NBODY_CONSISTENCY_ERROR;
}
/* Bodies must be set before trying to use this */
if (!st->bodytab)
{
mw_printf("Bodies not set for CL state initialization\n");
return NBODY_CONSISTENCY_ERROR;
}
if (ctx->potentialType == EXTERNAL_POTENTIAL_CUSTOM_LUA)
{
mw_printf("Cannot use Lua potential with OpenCL\n");
return NBODY_UNSUPPORTED;
}
devInfo = &st->ci->di;
if (!nbCheckDevCapabilities(devInfo, ctx, st->nbody))
return NBODY_CAPABILITY_ERROR;
if ( nbSetThreadCounts(st->workSizes, devInfo, ctx)
|| nbSetWorkSizes(st->workSizes, devInfo, st->nbody, st->ignoreResponsive))
return NBODY_ERROR;
st->effNBody = nbFindEffectiveNBody(st->workSizes, st->usesExact, st->nbody);
st->maxDepth = nbFindMaxDepthForDevice(devInfo, st->workSizes, ctx->useQuad);
st->usesConsistentMemory = (mwIsNvidiaGPUDevice(devInfo) && mwNvidiaInlinePTXAvailable(st->ci->plat))
|| mwDeviceHasConsistentMemory(devInfo);
if (nbLoadKernels(ctx, st))
return NBODY_CL_ERROR;
err = nbCreateBuffers(ctx, st);
if (err != CL_SUCCESS)
return NBODY_CL_ERROR;
err = nbSetInitialTreeStatus(st);
if (err != CL_SUCCESS)
return NBODY_CL_ERROR;
err = nbSetAllKernelArguments(st);
if (err != CL_SUCCESS)
return NBODY_CL_ERROR;
err = nbMarshalBodies(st, CL_TRUE);
if (err != CL_SUCCESS)
{
mw_printf("Error marshalling initial bodies\n");
return NBODY_CL_ERROR;
}
return NBODY_SUCCESS;
}