Primitive storage and static model improvements #171
Conversation
…cleanup in related places.
…primitive structure as the BSP.
…and primitive offset with separate VBO index and primitive offset stored in a UBO.
…ntents with shared structure declarations.
…n, massively refactored.
… models, instance those BLAS instead of going through the shader.
|
I've found some minor things (typos etc), but there's nothing major (...that I noticed). |
…eometry, that's confusing the path tracer.
|
Thanks for the review! All feedback has been addressed. |
…RF_TRANS33/66 flags and entity->alpha. Improved the path tracer logic to handle complex translucent geometry better and make it cast shadows.
- Replaced the arrays of primitive attributes that were specific to BSP or models with a single, unified storage format: VboPrimitive. - Added support for static models with their own BLAS that never go through the instnace_geometry shader. - Increased the limits for primitive and mesh counts, removed the limits on BSP or instanced buffer sizes. - Expanded support for translucent materials, so cl_gunalpha works now. - Fixed some long-standing resource leaks on map changes.
|
Merged through squash in e7d671d |
|
Somehow I missed this PR entirely in my inbox - I'm sorry! I'll go over the commits even though it's too late, haha EDIT: looks good, no issues other than the one I mentioned which is more of a style thing. |
| CHECK(iqmData->jointParents = (int*)MOD_Malloc(header->num_joints * sizeof(int))); | ||
| CHECK(iqmData->bindJoints = (float*)MOD_Malloc(header->num_joints * 12 * sizeof(float))); // bind joint matricies | ||
| CHECK(iqmData->invBindJoints = (float*)MOD_Malloc(header->num_joints * 12 * sizeof(float))); // inverse bind joint matricies | ||
| CHECK(iqmData->jointNames = MOD_Malloc(joint_names)); |
There was a problem hiding this comment.
this just reminded me, on my local branch I changed jointNames to be an array of char[MAX_QPATH] so that I can easily index the joint names at runtime. Currently, you need to loop over the string via strlen which is pretty inefficient.. not that it totally matters I guess since you shouldn't be looping through joint names every frame at runtime, but just a random note that I think a single long string of "name data" is a bit harder to parse through than just fixed-length string indices.
This PR may look a bit odd coming from me, but I just want to have some more eyes on the changes before I merge them... It's large. The changes will be merged with squash most likely.
@res2k @Paril specifically
Summary: