void ceDeviceGL20::DrawIndexPrimitives (cePrimitiveType pType, ceDataType iType, unsigned count, bool unbind)
{
assert(_vertexDeclaration);
assert(_indices);
for (unsigned i = 0, j = _vertexDeclaration->GetNumberOfStreams (); i < j; i++)
{
void* ptr = _vertexStreams[i]->Bind ();
_vertexDeclaration->BindStream (_program, i, ptr);
}
_indices->Bind ();
prim_type = convert_primitive_type (pType);
prim_count = count;
index_type = convert_data_type (iType);
glDrawElements (prim_type, prim_count, index_type, 0);
if (unbind)
{
for (unsigned i = 0, j = _vertexDeclaration->GetNumberOfStreams (); i < j; i++)
{
_vertexDeclaration->UnbindStream (_program, i);
}
}
}
/**
* \brief Helper function to check for restrictions on `gpudata` to be used in
* nccl
* collective operations.
*/
static inline int check_restrictions(gpudata *src, size_t offsrc,
gpudata *dest, size_t offdest,
size_t count, int typecode,
int opcode, gpucomm *comm,
ncclDataType_t *datatype,
ncclRedOp_t *op) {
size_t op_size;
// Check if count is larger than INT_MAX
// TODO remove whenif nccl adapts to size_t
if (count > INT_MAX)
return error_set(comm->ctx->err, GA_XLARGE_ERROR, "Count too large for int");
// src, dest and comm must refer to the same context
if (src->ctx != comm->ctx)
return error_set(comm->ctx->err, GA_VALUE_ERROR, "source and comm context differ");
if (dest != NULL && dest->ctx != comm->ctx)
return error_set(comm->ctx->err, GA_VALUE_ERROR, "destination and comm context differ");
// typecode must correspond to a valid ncclDataType_t
if (datatype != NULL) {
*datatype = convert_data_type(typecode);
if (*datatype == ncclNumTypes)
return error_set(comm->ctx->err, GA_INVALID_ERROR, "Invalid data type");
}
// opcode must correspond to a valid ncclRedOp_t
if (op != NULL) {
*op = convert_reduce_op(opcode);
if (*op == ncclNumOps)
return error_set(comm->ctx->err, GA_INVALID_ERROR, "Invalid reduce op");
}
// offsets must not be larger than gpudata's size itself
// (else out of alloc-ed mem scope)
assert(!(offsrc > src->sz));
assert(!(dest != NULL && offdest > dest->sz));
// size to operate upon must be able to fit inside the gpudata (incl offsets)
op_size = count * gpuarray_get_elsize(typecode);
if ((src->sz - offsrc) < op_size)
return error_set(comm->ctx->err, GA_VALUE_ERROR, "source too small for operation");
if (dest != NULL && (dest->sz - offdest) < op_size)
return error_set(comm->ctx->err, GA_VALUE_ERROR, "destination too small for operation");
return GA_NO_ERROR;
}
