Uniform data is now scoped to the command buffer. I also addressed some D3D11 implementation inconsistencies that should improve performance via the Don't Touch The Driver principle.
Went back to a pooling strategy for uniform buffers to address situations where lots of command buffers are created and submitted (for example, when there is backpressure in immediate/mailbox mode).
The number of uniform slots per stage is now 4 instead of 14.
Addresses #54
Uniform data is now scoped to the command buffer. I also addressed some D3D11 implementation inconsistencies that should improve performance via the Don't Touch The Driver principle.
Needs corresponding Metal changes.