MT encoder WIP.

src/bgfx.cpp

@@ -887,7 +887,7 @@ namespace bgfx
 		&&  0 == m_draw.m_numIndices)
 		{
 			discard();
-			++m_frame->m_numDropped;
+			++m_numDropped;
 			return;
 		}
 
@@ -895,11 +895,13 @@ namespace bgfx
 		if (BGFX_CONFIG_MAX_DRAW_CALLS-1 <= renderItemIdx)
 		{
 			discard();
-			++m_frame->m_numDropped;
+			++m_numDropped;
 			return;
 		}
 
-		m_frame->m_uniformEnd = m_frame->m_uniformBuffer->getPos();
+		++m_numSubmitted;
+
+		m_uniformEnd = m_uniformBuffer->getPos();
 
 		m_key.m_program = kInvalidHandle == _program.idx
 			? 0
@@ -922,8 +924,9 @@ namespace bgfx
 		m_frame->m_sortKeys[renderItemIdx]   = key;
 		m_frame->m_sortValues[renderItemIdx] = RenderItemCount(renderItemIdx);
 
-		m_draw.m_uniformBegin = m_frame->m_uniformBegin;
-		m_draw.m_uniformEnd   = m_frame->m_uniformEnd;
+		m_draw.m_uniformIdx   = m_uniformIdx;
+		m_draw.m_uniformBegin = m_uniformBegin;
+		m_draw.m_uniformEnd   = m_uniformEnd;
 		m_draw.m_stateFlags  |= m_stateFlags;
 
 		uint32_t numVertices = UINT32_MAX;
@@ -953,7 +956,7 @@ namespace bgfx
 		{
 			m_draw.clear();
 			m_bind.clear();
-			m_frame->m_uniformBegin = m_frame->m_uniformEnd;
+			m_uniformBegin = m_uniformEnd;
 			m_stateFlags = BGFX_STATE_NONE;
 		}
 	}
@@ -975,11 +978,13 @@ namespace bgfx
 		if (BGFX_CONFIG_MAX_DRAW_CALLS-1 <= renderItemIdx)
 		{
 			discard();
-			++m_frame->m_numDropped;
+			++m_numDropped;
 			return;
 		}
 
-		m_frame->m_uniformEnd = m_frame->m_uniformBuffer->getPos();
+		++m_numSubmitted;
+
+		m_uniformEnd = m_uniformBuffer->getPos();
 
 		m_compute.m_startMatrix = m_draw.m_startMatrix;
 		m_compute.m_numMatrices = m_draw.m_numMatrices;
@@ -997,14 +1002,15 @@ namespace bgfx
 		m_frame->m_sortKeys[renderItemIdx]   = key;
 		m_frame->m_sortValues[renderItemIdx] = RenderItemCount(renderItemIdx);
 
-		m_compute.m_uniformBegin = m_frame->m_uniformBegin;
-		m_compute.m_uniformEnd   = m_frame->m_uniformEnd;
+		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();
 		m_bind.clear();
-		m_frame->m_uniformBegin = m_frame->m_uniformEnd;
+		m_uniformBegin = m_uniformEnd;
 	}
 
 	void EncoderImpl::blit(uint8_t _id, TextureHandle _dst, uint8_t _dstMip, uint16_t _dstX, uint16_t _dstY, uint16_t _dstZ, TextureHandle _src, uint8_t _srcMip, uint16_t _srcX, uint16_t _srcY, uint16_t _srcZ, uint16_t _width, uint16_t _height, uint16_t _depth)
@@ -1437,7 +1443,7 @@ namespace bgfx
 
 		frameNoRenderWait();
 
-		m_encoder[0].begin(m_submit);
+		m_encoder[0].begin(m_submit, 0);
 		m_encoder0 = reinterpret_cast<Encoder*>(&m_encoder[0]);
 
 		// Make sure renderer init is called from render thread.
@@ -1691,9 +1697,9 @@ namespace bgfx
 				return NULL;
 			}
 
-			uint32_t idx = m_numEncoders++;
+			uint8_t idx = uint8_t(m_numEncoders++);
 			encoder = &m_encoder[idx];
-			encoder->begin(m_frame);
+			encoder->begin(m_submit, idx);
 		}
 #endif // BGFX_CONFIG_MULTITHREADED
 
@@ -1716,8 +1722,9 @@ namespace bgfx
 	uint32_t Context::frame(bool _capture)
 	{
 		bx::MutexScope resourceApiScope(m_resourceApiLock);
-		bx::MutexScope encoderApiScope(m_encoderApiLock);
+
 		encoderApiWait();
+		bx::MutexScope encoderApiScope(m_encoderApiLock);
 
 		m_encoder[0].end();
 		m_submit->m_capture = _capture;
@@ -1727,7 +1734,7 @@ namespace bgfx
 		renderSemWait();
 		frameNoRenderWait();
 
-		m_encoder[0].begin(m_submit);
+		m_encoder[0].begin(m_submit, 0);
 
 		return m_frames;
 	}

src/bgfx_p.h

@@ -1110,13 +1110,9 @@ namespace bgfx
 		uint32_t reserve(uint16_t* _num)
 		{
 			uint32_t num = *_num;
-			BX_WARN(m_num+num < BGFX_CONFIG_MAX_MATRIX_CACHE, "Matrix cache overflow. %d (max: %d)", m_num+num, BGFX_CONFIG_MAX_MATRIX_CACHE);
-			num = bx::uint32_min(num, BGFX_CONFIG_MAX_MATRIX_CACHE-m_num);
-			uint32_t first = m_num < BGFX_CONFIG_MAX_MATRIX_CACHE
-				? m_num
-				: BGFX_CONFIG_MAX_MATRIX_CACHE - 1
-				;
-			m_num += num;
+			uint32_t first = bx::atomicFetchAndAddsat<uint32_t>(&m_num, num, BGFX_CONFIG_MAX_MATRIX_CACHE - 1);
+			BX_WARN(first+num < BGFX_CONFIG_MAX_MATRIX_CACHE, "Matrix cache overflow. %d (max: %d)", first+num, BGFX_CONFIG_MAX_MATRIX_CACHE);
+			num = bx::uint32_min(num, BGFX_CONFIG_MAX_MATRIX_CACHE-1-first);
 			*_num = (uint16_t)num;
 			return first;
 		}
@@ -1165,17 +1161,16 @@ namespace bgfx
 
 		uint32_t add(uint16_t _x, uint16_t _y, uint16_t _width, uint16_t _height)
 		{
-			BX_CHECK(m_num+1 < BGFX_CONFIG_MAX_RECT_CACHE, "Rect cache overflow. %d (max: %d)", m_num, BGFX_CONFIG_MAX_RECT_CACHE);
+			const uint32_t first = bx::atomicFetchAndAddsat<uint32_t>(&m_num, 1, BGFX_CONFIG_MAX_RECT_CACHE-1);
+			BX_CHECK(first+1 < BGFX_CONFIG_MAX_RECT_CACHE, "Rect cache overflow. %d (max: %d)", first, BGFX_CONFIG_MAX_RECT_CACHE);
 
-			uint32_t first = m_num;
-			Rect& rect = m_cache[m_num];
+			Rect& rect = m_cache[first];
 
 			rect.m_x = _x;
 			rect.m_y = _y;
 			rect.m_width = _width;
 			rect.m_height = _height;
 
-			m_num++;
 			return first;
 		}
 
@@ -1460,6 +1455,7 @@ namespace bgfx
 			m_instanceDataBuffer.idx = kInvalidHandle;
 			m_indirectBuffer.idx     = kInvalidHandle;
 			m_occlusionQuery.idx     = kInvalidHandle;
+			m_uniformIdx = UINT8_MAX;
 		}
 
 		bool setStreamBit(uint8_t _stream, VertexBufferHandle _handle)
@@ -1490,6 +1486,7 @@ namespace bgfx
 		uint16_t m_scissor;
 		uint8_t  m_submitFlags;
 		uint8_t  m_streamMask;
+		uint8_t  m_uniformIdx;
 
 		IndexBufferHandle    m_indexBuffer;
 		VertexBufferHandle   m_instanceDataBuffer;
@@ -1509,6 +1506,7 @@ namespace bgfx
 			m_numZ         = 0;
 			m_numMatrices  = 0;
 			m_submitFlags  = BGFX_SUBMIT_EYE_FIRST;
+			m_uniformIdx = UINT8_MAX;
 
 			m_indirectBuffer.idx = kInvalidHandle;
 			m_startIndirect      = 0;
@@ -1527,6 +1525,7 @@ namespace bgfx
 		uint16_t m_numIndirect;
 		uint16_t m_numMatrices;
 		uint8_t  m_submitFlags;
+		uint8_t  m_uniformIdx;
 	};
 
 	union RenderItem
@@ -1697,8 +1696,7 @@ namespace bgfx
 	BX_ALIGN_DECL_CACHE_LINE(struct) Frame
 	{
 		Frame()
-			: m_uniformMax(0)
-			, m_waitSubmit(0)
+			: m_waitSubmit(0)
 			, m_waitRender(0)
 			, m_hmdInitialized(false)
 			, m_capture(false)
@@ -1717,7 +1715,11 @@ namespace bgfx
 
 		void create()
 		{
-			m_uniformBuffer = UniformBuffer::create();
+			for (uint32_t ii = 0; ii < BX_COUNTOF(m_uniformBuffer); ++ii)
+			{
+				m_uniformBuffer[ii] = UniformBuffer::create();
+			}
+
 			reset();
 			start();
 			m_textVideoMem = BX_NEW(g_allocator, TextVideoMem);
@@ -1725,7 +1727,10 @@ namespace bgfx
 
 		void destroy()
 		{
-			UniformBuffer::destroy(m_uniformBuffer);
+			for (uint32_t ii = 0; ii < BX_COUNTOF(m_uniformBuffer); ++ii)
+			{
+				UniformBuffer::destroy(m_uniformBuffer[ii]);
+			}
 			BX_DELETE(g_allocator, m_textVideoMem);
 		}
 
@@ -1738,17 +1743,13 @@ namespace bgfx
 
 		void start()
 		{
-			m_uniformBegin = 0;
-			m_uniformEnd   = 0;
 			m_frameCache.reset();
 			m_numRenderItems = 0;
-			m_numDropped     = 0;
 			m_numBlitItems   = 0;
 			m_iboffset = 0;
 			m_vboffset = 0;
 			m_cmdPre.start();
 			m_cmdPost.start();
-			m_uniformBuffer->reset();
 			m_capture = false;
 		}
 
@@ -1757,17 +1758,14 @@ namespace bgfx
 			m_cmdPre.finish();
 			m_cmdPost.finish();
 
-			m_uniformMax = bx::uint32_max(m_uniformMax, m_uniformBuffer->getPos() );
-			m_uniformBuffer->finish();
-
-			if (0 < m_numDropped)
-			{
-				BX_TRACE("Too many draw calls: %d, dropped %d (max: %d)"
-					, m_numRenderItems+m_numDropped
-					, m_numDropped
-					, BGFX_CONFIG_MAX_DRAW_CALLS
-					);
-			}
+//			if (0 < m_numDropped)
+//			{
+//				BX_TRACE("Too many draw calls: %d, dropped %d (max: %d)"
+//					, m_numRenderItems+m_numDropped
+//					, m_numDropped
+//					, BGFX_CONFIG_MAX_DRAW_CALLS
+//					);
+//			}
 		}
 
 		void sort();
@@ -1876,15 +1874,11 @@ namespace bgfx
 
 		uint32_t m_blitKeys[BGFX_CONFIG_MAX_BLIT_ITEMS+1];
 		BlitItem m_blitItem[BGFX_CONFIG_MAX_BLIT_ITEMS+1];
-		uint32_t m_uniformBegin;
-		uint32_t m_uniformEnd;
-		uint32_t m_uniformMax;
 
 		FrameCache m_frameCache;
-		UniformBuffer* m_uniformBuffer;
+		UniformBuffer* m_uniformBuffer[BGFX_CONFIG_MAX_ENCODERS];
 
 		uint32_t m_numRenderItems;
-		uint32_t m_numDropped;
 		uint16_t m_numBlitItems;
 
 		uint32_t m_iboffset;
@@ -1981,13 +1975,25 @@ namespace bgfx
 			discard();
 		}
 
-		void begin(Frame* _frame)
+		void begin(Frame* _frame, uint8_t _idx)
 		{
 			m_frame = _frame;
+
+			m_uniformIdx = _idx;
+			m_uniformBegin = 0;
+			m_uniformEnd   = 0;
+
+			m_uniformBuffer = m_frame->m_uniformBuffer[m_uniformIdx];
+			m_uniformBuffer->reset();
+
+			m_numSubmitted = 0;
+			m_numDropped   = 0;
 		}
 
 		void end()
 		{
+			m_uniformBuffer->finish();
+
 			if (BX_ENABLED(BGFX_CONFIG_DEBUG_OCCLUSION) )
 			{
 				m_occlusionQuerySet.clear();
@@ -2001,7 +2007,7 @@ namespace bgfx
 
 		void setMarker(const char* _name)
 		{
-			m_frame->m_uniformBuffer->writeMarker(_name);
+			m_uniformBuffer->writeMarker(_name);
 		}
 
 		void setUniform(UniformType::Enum _type, UniformHandle _handle, const void* _value, uint16_t _num)
@@ -2016,8 +2022,8 @@ namespace bgfx
 				m_uniformSet.insert(_handle.idx);
 			}
 
-			UniformBuffer::update(m_frame->m_uniformBuffer);
-			m_frame->m_uniformBuffer->writeUniform(_type, _handle.idx, _value, _num);
+			UniformBuffer::update(m_uniformBuffer);
+			m_uniformBuffer->writeUniform(_type, _handle.idx, _value, _num);
 		}
 
 		void setState(uint64_t _state, uint32_t _rgba)
@@ -2261,13 +2267,21 @@ namespace bgfx
 		RenderCompute m_compute;
 		RenderBind    m_bind;
 
+		uint32_t m_numSubmitted;
+		uint32_t m_numDropped;
+
+		uint32_t m_uniformBegin;
+		uint32_t m_uniformEnd;
 		uint64_t m_stateFlags;
 		uint32_t m_numVertices[BGFX_CONFIG_MAX_VERTEX_STREAMS];
+		uint8_t  m_uniformIdx;
 		bool     m_discard;
 
 		typedef stl::unordered_set<uint16_t> HandleSet;
 		HandleSet m_uniformSet;
 		HandleSet m_occlusionQuerySet;
+
+		UniformBuffer* m_uniformBuffer;
 	};
 
 	struct VertexDeclRef
@@ -4345,9 +4359,9 @@ namespace bgfx
 			}
 		}
 
-		void encoderApiWait()
+		BX_NO_INLINE void encoderApiWait()
 		{
-			for (uint32_t ii = 1; ii < m_numEncoders; ++ii)
+			for (uint32_t ii = 1, num = m_numEncoders; ii < num; ++ii)
 			{
 				m_encoderApiSem.wait();
 			}

src/renderer_d3d11.cpp

@@ -5810,7 +5810,7 @@ BX_PRAGMA_DIAGNOSTIC_POP();
 
 					bool programChanged = false;
 					bool constantsChanged = compute.m_uniformBegin < compute.m_uniformEnd;
-					rendererUpdateUniforms(this, _render->m_uniformBuffer, compute.m_uniformBegin, compute.m_uniformEnd);
+					rendererUpdateUniforms(this, _render->m_uniformBuffer[compute.m_uniformIdx], compute.m_uniformBegin, compute.m_uniformEnd);
 
 					if (key.m_program != programIdx)
 					{
@@ -6079,7 +6079,7 @@ BX_PRAGMA_DIAGNOSTIC_POP();
 
 				bool programChanged = false;
 				bool constantsChanged = draw.m_uniformBegin < draw.m_uniformEnd;
-				rendererUpdateUniforms(this, _render->m_uniformBuffer, draw.m_uniformBegin, draw.m_uniformEnd);
+				rendererUpdateUniforms(this, _render->m_uniformBuffer[draw.m_uniformIdx], draw.m_uniformBegin, draw.m_uniformEnd);
 
 				if (key.m_program != programIdx)
 				{
@@ -6598,7 +6598,7 @@ BX_PRAGMA_DIAGNOSTIC_POP();
 				}
 
 				tvm.printf(10, pos++, 0x8e, "      Indices: %7d ", statsNumIndices);
-				tvm.printf(10, pos++, 0x8e, " Uniform size: %7d, Max: %7d ", _render->m_uniformEnd, _render->m_uniformMax);
+//				tvm.printf(10, pos++, 0x8e, " Uniform size: %7d, Max: %7d ", _render->m_uniformEnd, _render->m_uniformMax);
 				tvm.printf(10, pos++, 0x8e, "     DVB size: %7d ", _render->m_vboffset);
 				tvm.printf(10, pos++, 0x8e, "     DIB size: %7d ", _render->m_iboffset);
 

src/renderer_d3d12.cpp

@@ -5318,7 +5318,7 @@ data.NumQualityLevels = 0;
 					if (compute.m_uniformBegin < compute.m_uniformEnd
 					||  currentProgramIdx != key.m_program)
 					{
-						rendererUpdateUniforms(this, _render->m_uniformBuffer, compute.m_uniformBegin, compute.m_uniformEnd);
+						rendererUpdateUniforms(this, _render->m_uniformBuffer[compute.m_uniformIdx], compute.m_uniformBegin, compute.m_uniformEnd);
 
 						currentProgramIdx = key.m_program;
 						ProgramD3D12& program = m_program[currentProgramIdx];
@@ -5432,7 +5432,7 @@ data.NumQualityLevels = 0;
 					primIndex = uint8_t(pt>>BGFX_STATE_PT_SHIFT);
 				}
 
-				rendererUpdateUniforms(this, _render->m_uniformBuffer, draw.m_uniformBegin, draw.m_uniformEnd);
+				rendererUpdateUniforms(this, _render->m_uniformBuffer[draw.m_uniformIdx], draw.m_uniformBegin, draw.m_uniformEnd);
 
 				if (isValid(draw.m_stream[0].m_handle) )
 				{
@@ -5898,7 +5898,7 @@ data.NumQualityLevels = 0;
 				}
 
 				tvm.printf(10, pos++, 0x8e, "      Indices: %7d ", statsNumIndices);
-				tvm.printf(10, pos++, 0x8e, " Uniform size: %7d, Max: %7d ", _render->m_uniformEnd, _render->m_uniformMax);
+//				tvm.printf(10, pos++, 0x8e, " Uniform size: %7d, Max: %7d ", _render->m_uniformEnd, _render->m_uniformMax);
 				tvm.printf(10, pos++, 0x8e, "     DVB size: %7d ", _render->m_vboffset);
 				tvm.printf(10, pos++, 0x8e, "     DIB size: %7d ", _render->m_iboffset);
 

src/renderer_d3d9.cpp

@@ -4083,7 +4083,7 @@ namespace bgfx { namespace d3d9
 
 				bool programChanged = false;
 				bool constantsChanged = draw.m_uniformBegin < draw.m_uniformEnd;
-				rendererUpdateUniforms(this, _render->m_uniformBuffer, draw.m_uniformBegin, draw.m_uniformEnd);
+				rendererUpdateUniforms(this, _render->m_uniformBuffer[draw.m_uniformIdx], draw.m_uniformBegin, draw.m_uniformEnd);
 
 				if (key.m_program != programIdx)
 				{
@@ -4444,7 +4444,7 @@ namespace bgfx { namespace d3d9
 				}
 
 				tvm.printf(10, pos++, 0x8e, "      Indices: %7d ", statsNumIndices);
-				tvm.printf(10, pos++, 0x8e, " Uniform size: %7d, Max: %7d ", _render->m_uniformEnd, _render->m_uniformMax);
+//				tvm.printf(10, pos++, 0x8e, " Uniform size: %7d, Max: %7d ", _render->m_uniformEnd, _render->m_uniformMax);
 				tvm.printf(10, pos++, 0x8e, "     DVB size: %7d ", _render->m_vboffset);
 				tvm.printf(10, pos++, 0x8e, "     DIB size: %7d ", _render->m_iboffset);
 

src/renderer_gl.cpp

@@ -6827,7 +6827,7 @@ namespace bgfx { namespace gl
 						if (0 != barrier)
 						{
 							bool constantsChanged = compute.m_uniformBegin < compute.m_uniformEnd;
-							rendererUpdateUniforms(this, _render->m_uniformBuffer, compute.m_uniformBegin, compute.m_uniformEnd);
+							rendererUpdateUniforms(this, _render->m_uniformBuffer[compute.m_uniformIdx], compute.m_uniformBegin, compute.m_uniformEnd);
 
 							if (constantsChanged
 							&&  NULL != program.m_constantBuffer)
@@ -7242,7 +7242,7 @@ namespace bgfx { namespace gl
 				bool programChanged = false;
 				bool constantsChanged = draw.m_uniformBegin < draw.m_uniformEnd;
 				bool bindAttribs = false;
-				rendererUpdateUniforms(this, _render->m_uniformBuffer, draw.m_uniformBegin, draw.m_uniformEnd);
+				rendererUpdateUniforms(this, _render->m_uniformBuffer[draw.m_uniformIdx], draw.m_uniformBegin, draw.m_uniformEnd);
 
 				if (key.m_program != programIdx)
 				{
@@ -7710,7 +7710,7 @@ namespace bgfx { namespace gl
 				}
 
 				tvm.printf(10, pos++, 0x8e, "      Indices: %7d ", statsNumIndices);
-				tvm.printf(10, pos++, 0x8e, " Uniform size: %7d, Max: %7d ", _render->m_uniformEnd, _render->m_uniformMax);
+//				tvm.printf(10, pos++, 0x8e, " Uniform size: %7d, Max: %7d ", _render->m_uniformEnd, _render->m_uniformMax);
 				tvm.printf(10, pos++, 0x8e, "     DVB size: %7d ", _render->m_vboffset);
 				tvm.printf(10, pos++, 0x8e, "     DIB size: %7d ", _render->m_iboffset);
 

src/renderer_vk.cpp

@@ -3961,7 +3961,7 @@ BX_UNUSED(currentSamplerStateIdx);
 					if (compute.m_uniformBegin < compute.m_uniformEnd
 					||  currentProgramIdx != key.m_program)
 					{
-						rendererUpdateUniforms(this, _render->m_uniformBuffer, compute.m_uniformBegin, compute.m_uniformEnd);
+						rendererUpdateUniforms(this, _render->m_uniformBuffer[compute.m_uniformIdx], compute.m_uniformBegin, compute.m_uniformEnd);
 
 						currentProgramIdx = key.m_program;
 						ProgramVK& program = m_program[currentProgramIdx];
@@ -4073,7 +4073,7 @@ BX_UNUSED(currentSamplerStateIdx);
 					primIndex = uint8_t(pt>>BGFX_STATE_PT_SHIFT);
 				}
 
-				rendererUpdateUniforms(this, _render->m_uniformBuffer, draw.m_uniformBegin, draw.m_uniformEnd);
+				rendererUpdateUniforms(this, _render->m_uniformBuffer[draw.m_uniformIdx], draw.m_uniformBegin, draw.m_uniformEnd);
 
 				if (isValid(draw.m_stream[0].m_handle) )
 				{
@@ -4529,7 +4529,7 @@ BX_UNUSED(presentMin, presentMax);
  				}
 
 				tvm.printf(10, pos++, 0x8e, "      Indices: %7d ", statsNumIndices);
-				tvm.printf(10, pos++, 0x8e, " Uniform size: %7d, Max: %7d ", _render->m_uniformEnd, _render->m_uniformMax);
+//				tvm.printf(10, pos++, 0x8e, " Uniform size: %7d, Max: %7d ", _render->m_uniformEnd, _render->m_uniformMax);
 				tvm.printf(10, pos++, 0x8e, "     DVB size: %7d ", _render->m_vboffset);
 				tvm.printf(10, pos++, 0x8e, "     DIB size: %7d ", _render->m_iboffset);