Metal: Occlusion query WIP.

src/renderer_mtl.h

@@ -712,6 +712,29 @@ namespace bgfx { namespace mtl
 		uint8_t m_num; // number of color handles
 	};
 
+	struct OcclusionQueryMTL
+	{
+		OcclusionQueryMTL()
+			: m_control(BX_COUNTOF(m_query) )
+		{
+		}
+
+		void postReset();
+		void preReset();
+		void begin(RenderCommandEncoder& _rce, Frame* _render, OcclusionQueryHandle _handle);
+		void end(RenderCommandEncoder& _rce);
+		void resolve(Frame* _render, bool _wait = false);
+
+		struct Query
+		{
+			OcclusionQueryHandle m_handle;
+		};
+
+		Buffer m_buffer;
+		Query m_query[BGFX_CONFIG_MAX_OCCUSION_QUERIES];
+		bx::RingBufferControl m_control;
+	};
+
 } /* namespace metal */ } // namespace bgfx
 
 #endif // BGFX_CONFIG_RENDERER_METAL

src/renderer_mtl.mm

@@ -348,8 +348,8 @@ namespace bgfx { namespace mtl
 
 			if (NULL != NSClassFromString(@"CAMetalLayer") )
 			{
-				//on iOS we need the layer as CAmetalLayer
 #if BX_PLATFORM_IOS
+				{
 					CAMetalLayer* metalLayer = (CAMetalLayer*)g_platformData.nwh;
 					if (NULL == metalLayer
 					|| ![metalLayer isKindOfClass:NSClassFromString(@"CAMetalLayer")])
@@ -357,14 +357,16 @@ namespace bgfx { namespace mtl
 						BX_WARN(NULL != m_device, "Unable to create Metal device. Please set platform data window to a CAMetalLayer");
 						return false;
 					}
-				m_metalLayer = metalLayer;
-#elif BX_PLATFORM_OSX
-				// create and set metalLayer
-				NSWindow* nsWindow = (NSWindow*)g_platformData.nwh;
 
+					m_metalLayer = metalLayer;
+				}
+#elif BX_PLATFORM_OSX
+				{
+					NSWindow* nsWindow = (NSWindow*)g_platformData.nwh;
 					[nsWindow.contentView setWantsLayer:YES];
 					m_metalLayer = [CAMetalLayer layer];
 					[nsWindow.contentView setLayer:m_metalLayer];
+				}
 #endif // BX_PLATFORM_*
 
 				m_device = (id<MTLDevice>)g_platformData.context;
@@ -401,8 +403,8 @@ namespace bgfx { namespace mtl
 			m_uniformBufferFragmentOffset = 0;
 
 			g_caps.supported |= (0
-								 | BGFX_CAPS_TEXTURE_COMPARE_LEQUAL
 								 | BGFX_CAPS_TEXTURE_3D
+								 | BGFX_CAPS_TEXTURE_COMPARE_LEQUAL
 								 | BGFX_CAPS_INSTANCING
 								 | BGFX_CAPS_VERTEX_ATTRIB_HALF
 //								 | BGFX_CAPS_FRAGMENT_DEPTH
@@ -410,6 +412,9 @@ namespace bgfx { namespace mtl
 								 | BGFX_CAPS_COMPUTE
 								 | BGFX_CAPS_INDEX32
 								 | BGFX_CAPS_DRAW_INDIRECT
+//								 | BGFX_CAPS_TEXTURE_BLIT
+//								 | BGFX_CAPS_TEXTURE_READ_BACK
+								 | BGFX_CAPS_OCCLUSION_QUERY
 								 );
 
 			g_caps.maxTextureSize = 2048; //ASK: real caps width/height: 4096, but max depth(3D) size is only: 2048
@@ -483,11 +488,15 @@ namespace bgfx { namespace mtl
 				bx::snprintf(s_viewName[ii], BGFX_CONFIG_MAX_VIEW_NAME_RESERVED+1, "%3d   ", ii);
 			}
 
+			m_occlusionQuery.preReset();
+
 			return true;
 		}
 
 		void shutdown()
 		{
+			m_occlusionQuery.postReset();
+
 			for (uint32_t ii = 0; ii < BX_COUNTOF(m_shaders); ++ii)
 			{
 				m_shaders[ii].destroy();
@@ -507,9 +516,10 @@ namespace bgfx { namespace mtl
 			MTL_RELEASE(m_samplerDescriptor);
 
 			MTL_RELEASE(m_backBufferDepth);
-#if BX_PLATFORM_IOS
+			if (BX_ENABLED(BX_PLATFORM_IOS) )
+			{
 				MTL_RELEASE(m_backBufferStencil);
-#endif // BX_PLATFORM_*
+			}
 
 			MTL_RELEASE(m_uniformBuffer);
 			MTL_RELEASE(m_commandQueue);
@@ -930,19 +940,7 @@ namespace bgfx { namespace mtl
 					release(m_backBufferDepth);
 				}
 				m_backBufferDepth   = m_device.newTextureWithDescriptor(m_textureDescriptor);
-
-#if 0
-			m_textureDescriptor.pixelFormat = MTLPixelFormatStencil8;
-
-			if (NULL != m_backBufferStencil)
-			{
-				release(m_backBufferStencil);
-			}
-
-			m_backBufferStencil = m_device.newTextureWithDescriptor(m_textureDescriptor);
-#else
 				m_backBufferStencil = m_backBufferDepth;
-#endif // 0
 
 				bx::HashMurmur2A murmur;
 				murmur.begin();
@@ -964,14 +962,12 @@ namespace bgfx { namespace mtl
 
 		void setShaderUniform(uint8_t _flags, uint32_t _loc, const void* _val, uint32_t _numRegs)
 		{
-		if (_flags&BGFX_UNIFORM_FRAGMENTBIT)
+			uint32_t offset = 0 != (_flags&BGFX_UNIFORM_FRAGMENTBIT)
-		{
+				? m_uniformBufferFragmentOffset
-			memcpy(&((char*)m_uniformBuffer.contents())[m_uniformBufferFragmentOffset + _loc], _val, _numRegs*16);
+				: m_uniformBufferVertexOffset
-		}
+				;
-		else
+			uint8_t* dst = (uint8_t*)m_uniformBuffer.contents();
-		{
+			memcpy(&dst[offset + _loc], _val, _numRegs*16);
-			memcpy(&((char*)m_uniformBuffer.contents())[m_uniformBufferVertexOffset + _loc], _val, _numRegs*16);
-		}
 		}
 
 		void setShaderUniform4f(uint8_t _flags, uint32_t _loc, const void* _val, uint32_t _numRegs)
@@ -1026,7 +1022,7 @@ namespace bgfx { namespace mtl
 				switch ( (uint32_t)type)
 				{
 				case UniformType::Mat3:
-				case UniformType::Mat3|BGFX_UNIFORM_FRAGMENTBIT: \
+				case UniformType::Mat3|BGFX_UNIFORM_FRAGMENTBIT:
 					{
 						float* value = (float*)data;
 						for (uint32_t ii = 0, count = num/3; ii < count; ++ii,  loc += 3*16, value += 9)
@@ -1066,7 +1062,7 @@ namespace bgfx { namespace mtl
 			}
 		}
 
-	void setFrameBuffer(mtl::RenderPassDescriptor renderPassDescriptor, FrameBufferHandle _fbh, bool _msaa = true)
+		void setFrameBuffer(RenderPassDescriptor renderPassDescriptor, FrameBufferHandle _fbh, bool _msaa = true)
 		{
 			if (!isValid(_fbh) )
 			{
@@ -1086,7 +1082,7 @@ namespace bgfx { namespace mtl
 					renderPassDescriptor.colorAttachments[ii].texture = texture.m_ptr;
 				}
 
-			if (isValid(frameBuffer.m_depthHandle))
+				if (isValid(frameBuffer.m_depthHandle) )
 				{
 					const TextureMtl& texture = m_textures[frameBuffer.m_depthHandle.idx];
 					renderPassDescriptor.depthAttachment.texture = texture.m_ptr;
@@ -1197,6 +1193,12 @@ namespace bgfx { namespace mtl
 			return sampler;
 		}
 
+		bool isVisible(Frame* _render, OcclusionQueryHandle _handle, bool _visible)
+		{
+			m_occlusionQuery.resolve(_render);
+			return _visible == (0 != _render->m_occlusion[_handle.idx]);
+		}
+
 		uint32_t getBufferWidth()
 		{
 			return m_backBufferDepth.width();
@@ -1215,6 +1217,8 @@ namespace bgfx { namespace mtl
 		uint32_t      m_backBufferPixelFormatHash;
 		uint32_t      m_maxAnisotropy;
 
+		OcclusionQueryMTL m_occlusionQuery;
+
 		Buffer   m_uniformBuffer; //todo: use a pool of this
 		uint32_t m_uniformBufferVertexOffset;
 		uint32_t m_uniformBufferFragmentOffset;
@@ -1255,7 +1259,7 @@ namespace bgfx { namespace mtl
 		id <CAMetalDrawable> m_drawable;
 		CommandBuffer m_commandBuffer;
 		RenderCommandEncoder m_renderCommandEncoder;
-};
+	};
 
 	static RendererContextMtl* s_renderMtl;
 
@@ -2102,6 +2106,48 @@ namespace bgfx { namespace mtl
 		return denseIdx;
 	}
 
+	void OcclusionQueryMTL::postReset()
+	{
+		MTL_RELEASE(m_buffer);
+	}
+
+	void OcclusionQueryMTL::preReset()
+	{
+		m_buffer = s_renderMtl->m_device.newBufferWithLength(BX_COUNTOF(m_query)*8, 0);
+	}
+
+	void OcclusionQueryMTL::begin(RenderCommandEncoder& _rce, Frame* _render, OcclusionQueryHandle _handle)
+	{
+		while (0 == m_control.reserve(1) )
+		{
+			resolve(_render, true);
+		}
+
+		Query& query = m_query[m_control.m_current];
+		query.m_handle = _handle;
+		_rce.setVisibilityResultMode(MTLVisibilityResultModeBoolean, _handle.idx);
+	}
+
+	void OcclusionQueryMTL::end(RenderCommandEncoder& _rce)
+	{
+		Query& query = m_query[m_control.m_current];
+		_rce.setVisibilityResultMode(MTLVisibilityResultModeDisabled, query.m_handle.idx);
+		m_control.commit(1);
+	}
+
+	void OcclusionQueryMTL::resolve(Frame* _render, bool _wait)
+	{
+		BX_UNUSED(_wait);
+		while (0 != m_control.available() )
+		{
+			Query& query = m_query[m_control.m_read];
+
+			uint64_t result = ( (uint64_t*)m_buffer.contents() )[query.m_handle.idx];
+			_render->m_occlusion[query.m_handle.idx] = 0 < result;
+			m_control.consume(1);
+		}
+	}
+
 	void RendererContextMtl::submit(Frame* _render, ClearQuad& _clearQuad, TextVideoMemBlitter& _textVideoMemBlitter) BX_OVERRIDE
 	{
 		BX_UNUSED(_clearQuad);
@@ -2179,7 +2225,7 @@ namespace bgfx { namespace mtl
 		PrimInfo prim = s_primInfo[primIndex];
 
 		ProgramMtl* currentProgram = NULL;
-		mtl::RenderCommandEncoder  rce;
+		RenderCommandEncoder rce;
 
 		bool wasCompute = false;
 		bool viewHasScissor = false;
@@ -2193,6 +2239,8 @@ namespace bgfx { namespace mtl
 		uint32_t statsNumIndices = 0;
 		uint32_t statsKeyType[2] = {};
 
+		m_occlusionQuery.resolve(_render);
+
 		if (0 == (_render->m_debug&BGFX_DEBUG_IFH) )
 		{
 			bool viewRestart = false;
@@ -2258,7 +2306,8 @@ namespace bgfx { namespace mtl
 					viewScissorRect = viewHasScissor ? scissorRect : viewState.m_rect;
 					Clear& clr = _render->m_clear[view];
 
-					mtl::RenderPassDescriptor renderPassDescriptor = newRenderPassDescriptor();
+					RenderPassDescriptor renderPassDescriptor = newRenderPassDescriptor();
+					renderPassDescriptor.visibilityResultBuffer = m_occlusionQuery.m_buffer;
 
 					//todo: check FB size
 					uint32_t width  = getBufferWidth();
@@ -2370,6 +2419,14 @@ namespace bgfx { namespace mtl
 
 				const RenderDraw& draw = renderItem.draw;
 
+				const bool hasOcclusionQuery = 0 != (draw.m_stateFlags & BGFX_STATE_INTERNAL_OCCLUSION_QUERY);
+				if (isValid(draw.m_occlusionQuery)
+				&&  !hasOcclusionQuery
+				&&  !isVisible(_render, draw.m_occlusionQuery, 0 != (draw.m_submitFlags&BGFX_SUBMIT_INTERNAL_OCCLUSION_VISIBLE) ) )
+				{
+					continue;
+				}
+
 				const uint64_t newFlags = draw.m_stateFlags;
 				uint64_t changedFlags = currentState.m_stateFlags ^ draw.m_stateFlags;
 				currentState.m_stateFlags = newFlags;
@@ -2484,12 +2541,12 @@ namespace bgfx { namespace mtl
 				bool constantsChanged = draw.m_constBegin < draw.m_constEnd;
 				rendererUpdateUniforms(this, _render->m_uniformBuffer, draw.m_constBegin, draw.m_constEnd);
 
-				if (key.m_program != programIdx ||
+				if (key.m_program != programIdx
-					(BGFX_STATE_BLEND_MASK|BGFX_STATE_BLEND_EQUATION_MASK|BGFX_STATE_ALPHA_WRITE|BGFX_STATE_RGB_WRITE|BGFX_STATE_BLEND_INDEPENDENT|BGFX_STATE_MSAA) & changedFlags ||
+				|| (BGFX_STATE_BLEND_MASK|BGFX_STATE_BLEND_EQUATION_MASK|BGFX_STATE_ALPHA_WRITE|BGFX_STATE_RGB_WRITE|BGFX_STATE_BLEND_INDEPENDENT|BGFX_STATE_MSAA) & changedFlags
-					currentState.m_vertexBuffer.idx != draw.m_vertexBuffer.idx ||
+				||  currentState.m_vertexBuffer.idx != draw.m_vertexBuffer.idx
-					currentState.m_vertexDecl.idx != draw.m_vertexDecl.idx ||
+				||  currentState.m_vertexDecl.idx != draw.m_vertexDecl.idx
-					currentState.m_instanceDataStride != draw.m_instanceDataStride ||
+				||  currentState.m_instanceDataStride != draw.m_instanceDataStride
-					( (blendFactor != draw.m_rgba) && !!(newFlags & BGFX_STATE_BLEND_INDEPENDENT) ) )
+				|| ( (blendFactor != draw.m_rgba) && !!(newFlags & BGFX_STATE_BLEND_INDEPENDENT) ) )
 				{
 					programIdx = key.m_program;
 					currentState.m_vertexDecl         = draw.m_vertexDecl;
@@ -2532,14 +2589,14 @@ namespace bgfx { namespace mtl
 					uint32_t vertexUniformBufferSize   = program.m_vshConstantBufferSize;
 					uint32_t fragmentUniformBufferSize = program.m_fshConstantBufferSize;
 
-					if (vertexUniformBufferSize )
+					if (vertexUniformBufferSize)
 					{
 						m_uniformBufferVertexOffset = BX_ALIGN_MASK(m_uniformBufferVertexOffset, program.m_vshConstantBufferAlignmentMask);
 						rce.setVertexBuffer(m_uniformBuffer, m_uniformBufferVertexOffset, 0);
 					}
 
 					m_uniformBufferFragmentOffset = m_uniformBufferVertexOffset + vertexUniformBufferSize;
-					if (fragmentUniformBufferSize )
+					if (fragmentUniformBufferSize)
 					{
 						m_uniformBufferFragmentOffset = BX_ALIGN_MASK(m_uniformBufferFragmentOffset, program.m_fshConstantBufferAlignmentMask);
 						rce.setFragmentBuffer(m_uniformBuffer, m_uniformBufferFragmentOffset, 0);
@@ -2591,8 +2648,7 @@ namespace bgfx { namespace mtl
 				if (currentState.m_vertexBuffer.idx       != draw.m_vertexBuffer.idx
 				||  currentState.m_startVertex            != draw.m_startVertex
 				||  currentState.m_instanceDataBuffer.idx != draw.m_instanceDataBuffer.idx
-				||  currentState.m_instanceDataOffset     != draw.m_instanceDataOffset
+				||  currentState.m_instanceDataOffset     != draw.m_instanceDataOffset)
-				)
 				{
 					currentState.m_vertexBuffer           = draw.m_vertexBuffer;
 					currentState.m_startVertex            = draw.m_startVertex;
@@ -2635,6 +2691,11 @@ namespace bgfx { namespace mtl
 					uint32_t numPrimsRendered  = 0;
 					uint32_t numDrawIndirect   = 0;
 
+					if (hasOcclusionQuery)
+					{
+						m_occlusionQuery.begin(rce, _render, draw.m_occlusionQuery);
+					}
+
 					if (isValid(draw.m_indirectBuffer) )
 					{
 						 // TODO: indirect draw
@@ -2677,6 +2738,11 @@ namespace bgfx { namespace mtl
 						}
 					}
 
+					if (hasOcclusionQuery)
+					{
+						m_occlusionQuery.end(rce);
+					}
+
 					statsNumPrimsSubmitted[primIndex] += numPrimsSubmitted;
 					statsNumPrimsRendered[primIndex]  += numPrimsRendered;
 					statsNumInstances[primIndex]      += numInstances;