Fix: Fix Uniform buffer doing too many iterations when updating renderer uniforms

[LittleCodingFox]

When rendering 1275 drawcalls with heavy uniform use without discarding textures and uniforms, I was having terrible performance, worse than when discarding.

While investigating, I realized that when BGFX calls rendererUpdateUniforms, it always had _begin set to 0.

I looked at how EncoderImpl::dispatch always sets the m_uniformBegin to m_uniformEnd after each item is added, so I did the same for EncoderImpl::submit, and the issue is fixed! I actually have much better performance than when using discards, which is to be expected!

As requested, I tested the examples and they seem to work fine, so I believe there should be no side effects!

[bkaradzic]

Modify existing example from repo to demonstrate the problem you're having.

[LittleCodingFox]

I modified the hextile sample like so:

• Make it render the terrain 10000 times with BGFX_DISCARD_NONE

Tested in debug to prevent optimizations

Before my fix:

After my fix:

[bkaradzic]

I modified the hextile sample like so:

Do diff or paste code here so that I can use exactly what you're using, instead creating something that might not match what you have.

[LittleCodingFox]

Replace line 286 of hextile.cpp with:

				for (int i = 0; i < 10000; i++)
				{
					bgfx::submit(0, m_hextileProgram, 0, BGFX_DISCARD_NONE);
				}

[LittleCodingFox]

Also, these changes are essentially a 1:1 to how CS already did things:

(bgfx.cpp:1359)

		uint64_t key = m_key.encodeCompute();
		m_frame->m_sortKeys[renderItemIdx]   = key;
		m_frame->m_sortValues[renderItemIdx] = RenderItemCount(renderItemIdx);

		m_compute.m_uniformIdx   = m_uniformIdx;
		m_compute.m_uniformBegin = m_uniformBegin;
		m_compute.m_uniformEnd   = m_uniformEnd;
		m_frame->m_renderItem[renderItemIdx].compute = m_compute;
		m_frame->m_renderItemBind[renderItemIdx]     = m_bind;

		m_compute.clear(_flags);
		m_bind.clear(_flags);
		m_uniformBegin = m_uniformEnd;