Blender Git Loki

cleanup: GPUBufferCopyFunc defined twice

clean up GPUx vertex buffer API (part 1)

Removed unprimed version of GPUx_vertex_buffer_use function. The new
pattern is create, fill, prime, use {draw} done_using, use {draw}
done_using ...

Renamed GPUx_vertex_buffer_use_primed to GPUx_vertex_buffer_use since
there's only one version now.

Also removed the _primed suffix from GPUx_element_list_use.

TODO: make a wiki page about how to use this API

fix mismatched memory free

This string is copied with strdup during GPUx_specify_attrib, needs a
plain free.

new function: bind all attrib locations

Was binding these by hand in wireframe shader tests, but retyping the
name strings is a potential source of errors. Call this function instead
after creating VertexBuffer but before linking the GLSL program.

Revert "Add missing stubs for buffer generation."

This reverts commit 4040f0af0017409fad4912605527900ba6919173.

Merge branch 'master' into GPU_data_request

Add missing stubs for buffer generation.

Compilation fixes for C90 - still getting link errors related to buffer
generation routines.

WIP: fancy wireframe display

Exploring some possibilities with wireframe rendering beyond solid
color.

New effect: color edges based on angle between the faces joined. Loose
edges, perimeter edges, and edges joining 3+ faces are colored
differently.

Some of this code will move to Workflow Shaders. Other parts will move
to DerivedMesh data requests �?? AdjacentFaces (struct) and/or angle made
by adjacent faces (float).

The (incomplete) shader version uses VertexBuffer�??s generic attribute
API, which we want to switch everything to eventually. There is a GLSL
vertex shader *not* included in this commit because I don�??t know where
to put it�?�

The CPU version is temporary �?? it�??s equivalent to the shader setup +
matching GLSL shader.

This particular effect is view independent, so can be done CPU-side
once and drawn many times. Other effects I�??m working on are view
dependent, so really need the setup + shader split.

check extensions before use in GPUx

Using VAOs and integer vertex attributes.

VBO support is *not* checked because it�??s guaranteed to be there in a
GL 2.1 context.

VAOs are not part of GL 2.1, but enjoy nearly universal support.

gpu_shader4 is widely supported on recent hardware, not always on older
HW.

add runtime check for gpu_shader4 support

This extension includes a number of GLSL enhancements:
- full integer support
- interpolation qualifiers (flat, smooth, noperspective)
- gl_VertexID, gl_PrimitiveID
- custom fragment shader outputs, for MRT

single alloc per VertexBuffer, down from 2

Using flexible array member from C99.

Tested on LLVM/clang, also supported in GCC and MSVC according to docs:

https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
https://msdn.microsoft.com/en-us/library/b6fae073.aspx

minor cleanup

Mostly moving stuff to where it belongs.

UNUSED_VARS macro explains intent better than (void)var, so use it.

Considered replacing calloc with malloc for ElementList.indices, but
it�??s safer to zero this in case not all primitives are set. So calloc
it stays!

keep vertex data in main mem or VRAM (not both)

If we�??re using VBOs, free our own copy once GL has the data.

Data always lives in main memory while being constructed, so that we
don�??t need to call any OpenGL functions until it�??s time to draw.

minor cleanup

reset shading to SMOOTH or FLAT

missed this as part of GPUx interpolation commit

cleanup my own recent stuff

reduce VRAM usage for flat normals

Reduce number of GL verts from (3 * tess_tri_ct) to (1 * loop_ct).
Attribute data formats unchanged.

stock Suzanne mesh
507 verts, 1005 edges, 500 polys, 500 faces, 1968 loops

mode = NORMAL_DRAW_FLAT

(3 * tri_ct) verts, vec3s normals (+padding)
GPUx batch size: 63886 bytes

(3 * tri_ct) verts, vec3s normals (no padding)
GPUx batch size: 58080 bytes

loop_ct verts, vec3s normals (+padding)
GPUx batch size: 45166 bytes

loop_ct verts, vec3s normals (no padding)
GPUx batch size: 41232 bytes

reduce VRAM usage for loop normals

Use GL_SHORT instead of GL_FLOAT for normal attribute. Savings scale
proportional to loop count.

stock Suzanne mesh
507 verts, 1005 edges, 500 polys, 500 faces, 1968 loops

mode = NORMAL_DRAW_LOOP

vec3f normals
GPUx batch size: 53040 bytes

vec3s normals (+padding)
GPUx batch size: 45166 bytes

vec3s normals (no padding)
GPUx batch size: 41232 bytes

determine normal draw mode better

based on auto smooth, how many faces are flat vs. smooth, and viewport
overrides.