On Wed, 30 Jun 2021 at 22:07, Zebediah Figura (she/her) zfigura@codeweavers.com wrote:
On 6/30/21 2:46 PM, Henri Verbeet wrote:
Alternatively, we could do something like this in wined3d_device_context_unmap():
... if (context->ops->get_upload_bo(...)) { ... wined3d_device_context_upload_bo(...); return WINED3D_OK; } ... return wined3d_device_context_emit_unmap(...);Where I'm guessing wined3d_device_context_emit_unmap() contains pretty much the entirety of wined3d_cs_unmap()?
Right.
That had occurred to me as well, although I didn't implement it. I probably would leave it for a separate patch, though I'm not opposed to rolling it into this one.
Sure, we could do that in a subsequent patch.
which should also work, and may be preferable, because we could implement wined3d_device_context_map() and wined3d_device_context_update_sub_resource() in a similar way. I.e., I think we can make this work with either prepare_upload_bo()/get_upload_bo() or update_sub_resource()/map()/unmap() in the device context ops, but we probably don't need both sets of ops.
I'd agree the second approach outlined in this email looks clearer. That, or to just use update_sub_resource/map/unmap. I don't have a preference between those two, so I'll leave that decision to the maintainer ;-)
I'm leaning towards prepare_upload_bo()/get_upload_bo() because it's the smaller device context ops interface, and results in fewer levels of map/unmap API, but it's not a strong preference.