Use 17 bits for vertex id but substract first vtx index
Use 13 bits for poly number instead of 12
Panzer Front uses more than 4095 tr polys.
Issue #616
elan: Get rid of geometry shader and do geometry clipping manually.
pvr: Aggressively merge transparent polys in per-triangl and OIT mode, duplicating the last vertex to merge opposite cull mode strips.
gl4: reset OIT pointers in final shader. No need for clear pass.
gl4: alternate between 2 buffers for verts.index/tr poly params.
gl4: use gl_VertexID + poly number for poly index.
gl: new GlBuffer class.
Use w=1 to avoid multiply/divide of xy as in dx9
Disable rendering when video output disabled
Use original w value in pixel struct to avoid precision loss (samsptk)
vk oit: missing memory barrier was producing Tr modvol glitches (JSR)
gl oit: Tr modvol needs full vertex shader
stencil must be reset to zero if poly isn't shadowed -> fixes jet set
radio "see-through" shadows
take culling into account for modvols
oit: reduce per-pixel and per-poly mem usage
oit: framebuffers weren't resized when needed
oit: poly with depth write disabled weren't drawn (Soul calibur missing
water surface)
Fix glsl error when two-volume mode needs fog attachment
BufferData was always allocated in host visible mem, regardless of the
mem property argument.
NVidia gpus don't support blitting to 565 and 5551, which is needed for
mipmaps, so convert to 8888 instead.
Use optimal tiling for oit pointers image (linear not supported by
nvidia and probably slower anyway)
oit screen drawer uses multiple final attachments
Set non-oit depth attachment as transient and lazily allocate mem
flyinghead