Fix SSAO broken MSAA support

Fixes #476 Depth Downsampling needed a special shader to handle MSAA. SSAO_HS_ps shader was needing a special shader to handle MSAA due to gBuf_normals being MSAA. Rather than adding a special variant via ifdef, just disable explicit_resolve since the explicit MSAA surface is not needed in the sample. Reworked SSAO compositor to be applied in two stages, so that MSAA can be handled properly
OGRECave · Nov 10, 2024 · 93c3add · 93c3add
1 parent b262400
commit 93c3add
Show file tree

Hide file tree

Showing 6 changed files with 229 additions and 19 deletions.
diff --git a/Samples/2.0/Tutorials/Tutorial_SSAO/Tutorial_SSAO.cpp b/Samples/2.0/Tutorials/Tutorial_SSAO/Tutorial_SSAO.cpp
@@ -32,7 +32,10 @@ namespace Demo
         {
             Ogre::CompositorManager2 *compositorManager = mRoot->getCompositorManager2();
             return compositorManager->addWorkspace( mSceneManager, mRenderWindow->getTexture(), mCamera,
-                                                    "SSAOWorkspace", true );
+                                                    mRenderWindow->getSampleDescription().isMultisample()
+                                                        ? "SSAOWorkspaceMSAA"
+                                                        : "SSAOWorkspace",
+                                                    true );
         }
 
         void setupResources() override

diff --git a/Samples/Media/2.0/scripts/materials/Common/DepthUtils.material b/Samples/Media/2.0/scripts/materials/Common/DepthUtils.material
@@ -65,3 +65,71 @@ material Ogre/Depth/DownscaleMax
 		}
 	}
 }
+
+
+fragment_program Ogre/Depth/DownscaleMax_Subsample0_ps_GLSL glsl
+{
+	source DepthDownscaleMax_Subsample0_ps.glsl
+	default_params
+	{
+		param_named depthTexture int 0
+	}
+}
+
+fragment_program Ogre/Depth/DownscaleMax_Subsample0_ps_VK glslvk
+{
+	source DepthDownscaleMax_Subsample0_ps.glsl
+}
+
+fragment_program Ogre/Depth/DownscaleMax_Subsample0_ps_HLSL hlsl
+{
+	source DepthDownscaleMax_Subsample0_ps.hlsl
+	entry_point main
+	target ps_5_0 ps_4_0 ps_4_0_level_9_1 ps_4_0_level_9_3
+}
+
+fragment_program Ogre/Depth/DownscaleMax_Subsample0_ps_Metal metal
+{
+	source DepthDownscaleMax_Subsample0_ps.metal
+	shader_reflection_pair_hint Ogre/Compositor/Quad_vs
+}
+
+fragment_program Ogre/Depth/DownscaleMax_Subsample0_ps unified
+{
+	delegate Ogre/Depth/DownscaleMax_Subsample0_ps_GLSL
+	delegate Ogre/Depth/DownscaleMax_Subsample0_ps_VK
+	delegate Ogre/Depth/DownscaleMax_Subsample0_ps_HLSL
+	delegate Ogre/Depth/DownscaleMax_Subsample0_ps_Metal
+}
+
+// Downscales resolution of input depth texture by half (w/2 x h/2)
+// using a max filter (max depth of all 4 neighbours), MSAA version that reads only subsample 0.
+material Ogre/Depth/DownscaleMax_Subsample0
+{
+	technique
+	{
+		pass
+		{
+			depth_check on
+			depth_write on
+
+			depth_func always_pass
+
+			cull_hardware none
+
+			vertex_program_ref Ogre/Compositor/Quad_vs
+			{
+			}
+
+			fragment_program_ref Ogre/Depth/DownscaleMax_Subsample0_ps
+			{
+			}
+
+			texture_unit depthTexture
+			{
+				filtering			none
+				tex_address_mode	clamp
+			}
+		}
+	}
+}
diff --git a/Samples/Media/2.0/scripts/materials/Common/GLSL/DepthDownscaleMax_Subsample0_ps.glsl b/Samples/Media/2.0/scripts/materials/Common/GLSL/DepthDownscaleMax_Subsample0_ps.glsl
@@ -0,0 +1,17 @@
+#version ogre_glsl_ver_330
+
+vulkan_layout( ogre_t0 ) uniform texture2DMS depthTexture;
+
+in vec4 gl_FragCoord;
+//out float gl_FragDepth;
+
+void main()
+{
+	float fDepth0 = texelFetch( depthTexture, ivec2(gl_FragCoord.xy * 2.0), 0 ).x;
+	float fDepth1 = texelFetch( depthTexture, ivec2(gl_FragCoord.xy * 2.0) + ivec2( 0, 1 ), 0 ).x;
+	float fDepth2 = texelFetch( depthTexture, ivec2(gl_FragCoord.xy * 2.0) + ivec2( 1, 0 ), 0 ).x;
+	float fDepth3 = texelFetch( depthTexture, ivec2(gl_FragCoord.xy * 2.0) + ivec2( 1, 1 ), 0 ).x;
+
+	//gl_FragDepth = texelFetch( depthTexture, ivec2(gl_FragCoord.xy * 2.0), 0 ).x;
+	gl_FragDepth = max( max( fDepth0, fDepth1 ), max( fDepth2, fDepth3 ) );
+}
diff --git a/Samples/Media/2.0/scripts/materials/Common/HLSL/DepthDownscaleMax_Subsample0_ps.hlsl b/Samples/Media/2.0/scripts/materials/Common/HLSL/DepthDownscaleMax_Subsample0_ps.hlsl
@@ -0,0 +1,21 @@
+struct PS_INPUT
+{
+	float2 uv0			: TEXCOORD0;
+};
+
+Texture2DMS<float> depthTexture	: register(t0);
+
+float main
+(
+	PS_INPUT inPs,
+	float4 gl_FragCoord : SV_Position
+) : SV_Depth
+{
+	float fDepth0 = depthTexture.Load( int2(gl_FragCoord.xy * 2.0), 0 ).x;
+	float fDepth1 = depthTexture.Load( int2(gl_FragCoord.xy * 2.0) + int2( 0, 1 ), 0 ).x;
+	float fDepth2 = depthTexture.Load( int2(gl_FragCoord.xy * 2.0) + int2( 1, 0 ), 0 ).x;
+	float fDepth3 = depthTexture.Load( int2(gl_FragCoord.xy * 2.0) + int2( 1, 1 ), 0 ).x;
+
+	//return depthTexture.Load( int2(gl_FragCoord.xy * 2.0), 0 ) ).x;
+	return max( max( fDepth0, fDepth1 ), max( fDepth2, fDepth3 ) );
+}
diff --git a/Samples/Media/2.0/scripts/materials/Common/Metal/DepthDownscaleMax_Subsample0_ps.metal b/Samples/Media/2.0/scripts/materials/Common/Metal/DepthDownscaleMax_Subsample0_ps.metal
@@ -0,0 +1,33 @@
+#include <metal_stdlib>
+using namespace metal;
+
+struct PS_INPUT
+{
+	float2 uv0;
+};
+
+struct PS_OUTPUT
+{
+	float depth [[depth(any)]];
+};
+
+fragment PS_OUTPUT main_metal
+(
+	PS_INPUT inPs [[stage_in]],
+
+    depth2d_ms<float, access::read> depthTexture [[texture(0)]],
+
+	float4 gl_FragCoord [[position]]
+)
+{
+	uint2 iFragCoord = (uint2)(gl_FragCoord.xy * 2.0);
+    float fDepth0 = depthTexture.read( iFragCoord, 0 );
+    float fDepth1 = depthTexture.read( iFragCoord + uint2( 0, 1 ), 0 );
+    float fDepth2 = depthTexture.read( iFragCoord + uint2( 1, 0 ), 0 );
+    float fDepth3 = depthTexture.read( iFragCoord + uint2( 1, 1 ), 0 );
+
+	PS_OUTPUT outPs;
+	//outPs.depth =depthTexture.read( uint2(gl_FragCoord.xy * 2.0) ).x;
+	outPs.depth = max( max( fDepth0, fDepth1 ), max( fDepth2, fDepth3 ) );
+	return outPs;
+}
diff --git a/Samples/Media/2.0/scripts/materials/Tutorial_SSAO/SSAO_HS.compositor b/Samples/Media/2.0/scripts/materials/Tutorial_SSAO/SSAO_HS.compositor
@@ -1,17 +1,9 @@
-compositor_node SSAO_RenderNode
+compositor_node SSAO_MainRender
 {
-	in 0 rt_renderwindow
-
 	texture RT0				target_width target_height PFG_RGBA8_UNORM_SRGB		msaa_auto
-	texture gBufferNormals	target_width target_height PFG_R10G10B10A2_UNORM	msaa_auto explicit_resolve
-
-	texture depthTexture		target_width			target_height				PFG_D32_FLOAT msaa_auto
-	texture depthTextureCopy	target_width_scaled 0.5	target_height_scaled 0.5	PFG_D32_FLOAT
+	texture gBufferNormals	target_width target_height PFG_R10G10B10A2_UNORM	msaa_auto
 
-	texture ssaoTexture target_width_scaled 0.5 target_height_scaled 0.5 PFG_R16_FLOAT depth_pool 0
-
-	texture blurTextureHorizontal	target_width target_height PFG_R16_FLOAT depth_pool 0
-	texture blurTextureVertical		target_width target_height PFG_R16_FLOAT depth_pool 0
+	texture depthTexture	target_width			target_height				PFG_D32_FLOAT msaa_auto
 
 	rtv RT0
 	{
@@ -35,17 +27,63 @@ compositor_node SSAO_RenderNode
 			gen_normals_gbuffer true
 		}
 	}
-
+
+	out 0 RT0
+	out 1 gBufferNormals
+	out 2 depthTexture
+}
+
+compositor_node SSAO_DepthDownsampler
+{
+	in 0 depthTexture
+
+	texture depthTextureCopy	target_width_scaled 0.5	target_height_scaled 0.5	PFG_D32_FLOAT
+
 	target depthTextureCopy
 	{
 		pass render_quad
 		{
 			load { all dont_care }
 			material Ogre/Depth/DownscaleMax
-	    	input 0 depthTexture
+			input 0 depthTexture
+		}
+	}
+
+	out 0 depthTextureCopy
+}
+
+compositor_node SSAO_DepthDownsamplerMSAA
+{
+	in 0 depthTexture
+
+	texture depthTextureCopy	target_width_scaled 0.5	target_height_scaled 0.5	PFG_D32_FLOAT
+
+	target depthTextureCopy
+	{
+		pass render_quad
+		{
+			load { all dont_care }
+			material Ogre/Depth/DownscaleMax_Subsample0
+			input 0 depthTexture
 		}
 	}
 
+	out 0 depthTextureCopy
+}
+
+compositor_node SSAO_GenerationSSAO_and_Composite
+{
+	in 0 RT0
+	in 1 gBufferNormals
+	in 2 depthTextureCopy
+
+	in 3 rt_renderwindow
+
+	texture ssaoTexture target_width_scaled 0.5 target_height_scaled 0.5 PFG_R16_FLOAT depth_pool 0
+
+	texture blurTextureHorizontal	target_width target_height PFG_R16_FLOAT depth_pool 0
+	texture blurTextureVertical		target_width target_height PFG_R16_FLOAT depth_pool 0
+
 	target ssaoTexture
 	{
 		pass render_quad
@@ -56,13 +94,13 @@ compositor_node SSAO_RenderNode
 				clear_colour	1 1 1 1
 			}
 			material SSAO/HS
-	    	input 0 depthTextureCopy
+			input 0 depthTextureCopy
 			input 1 gBufferNormals
 
 			quad_normals	camera_far_corners_view_space
 		}
 	}
-	
+
 	target blurTextureHorizontal
 	{
 		pass render_quad
@@ -73,7 +111,7 @@ compositor_node SSAO_RenderNode
 			input 1 depthTextureCopy
 		}
 	}
-	
+
 	target blurTextureVertical
 	{
 		pass render_quad
@@ -94,7 +132,7 @@ compositor_node SSAO_RenderNode
 			input 0 blurTextureVertical
 			input 1 RT0
 		}
-		
+
 		pass render_scene
 		{
 			lod_update_list	off
@@ -109,5 +147,35 @@ compositor_node SSAO_RenderNode
 
 workspace SSAOWorkspace
 {
-	connect_output SSAO_RenderNode 0
+	/**
+		SSAO_MainRender							SSAO_GenerationSSAO_and_Composite
+			0 ---------------------------------> 0
+			1 ---------------------------------> 1
+					SSAO_DepthDownsampler
+			2 ---->  0  -----------------------> 2
+
+		connect_output ------------------------> 3
+	*/
+	connect SSAO_MainRender				2	SSAO_DepthDownsampler				0
+	connect SSAO_MainRender				0 1 SSAO_GenerationSSAO_and_Composite	0 1
+	connect SSAO_DepthDownsampler		0	SSAO_GenerationSSAO_and_Composite	2
+	connect_output							SSAO_GenerationSSAO_and_Composite	3
+}
+
+
+workspace SSAOWorkspaceMSAA
+{
+	/**
+		SSAO_MainRender							SSAO_GenerationSSAO_and_Composite
+			0 ---------------------------------> 0
+			1 ---------------------------------> 1
+					SSAO_DepthDownsamplerMSAA
+			2 ---->  0  -----------------------> 2
+
+		connect_output ------------------------> 3
+	*/
+	connect SSAO_MainRender				2	SSAO_DepthDownsamplerMSAA			0
+	connect SSAO_MainRender				0 1 SSAO_GenerationSSAO_and_Composite	0 1
+	connect SSAO_DepthDownsamplerMSAA	0	SSAO_GenerationSSAO_and_Composite	2
+	connect_output							SSAO_GenerationSSAO_and_Composite	3
 }