GS/HW: ROV support.

bin/resources/shaders/dx11/tfx.fx

@@ -50,6 +50,12 @@
 #define PS_AA1_TRIANGLE_SW_Z 3
 #endif
 
+#ifndef PS_ROV_DEPTH_NONE
+#define PS_ROV_DEPTH_NONE 0
+#define PS_ROV_DEPTH_READ_WRITE 1
+#define PS_ROV_DEPTH_READ_ONLY 2
+#endif
+
 #ifndef PS_FST
 #define PS_IIP 0
 #define PS_FST 0
@@ -111,6 +117,8 @@
 #define PS_TEX_IS_FB 0
 #define PS_AA1 0
 #define PS_ABE 0
+#define PS_ROV_COLOR 0
+#define PS_ROV_DEPTH 0
 #endif
 
 #ifndef VS_EXPAND_NONE
@@ -132,6 +140,12 @@
 #define SW_DEPTH (NEEDS_DEPTH_FOR_AFAIL || NEEDS_DEPTH_FOR_ZTST || NEEDS_DEPTH_FOR_AA1)
 #define ZWRITE (PS_ZFLOOR || PS_ZCLAMP || SW_DEPTH)
 
+#define PS_RETURN_COLOR_ROV (!PS_NO_COLOR && PS_ROV_COLOR)
+#define PS_RETURN_COLOR (!PS_NO_COLOR && !PS_ROV_COLOR)
+#define PS_RETURN_DEPTH_ROV (PS_ROV_DEPTH == PS_ROV_DEPTH_READ_WRITE)
+#define PS_RETURN_DEPTH (ZWRITE && !PS_ROV_DEPTH)
+#define PS_ROV_EARLYDEPTHSTENCIL (PS_ROV_COLOR && !PS_ROV_DEPTH && !ZWRITE)
+
 struct VS_INPUT
 {
 	float2 st : TEXCOORD0;
@@ -178,22 +192,27 @@ struct PS_INPUT
 
 #ifdef PIXEL_SHADER
 
-struct PS_OUTPUT
+struct PS_OUTPUT_REAL
 {
 #define NUM_RTS 0
-#if !PS_NO_COLOR
-#if PS_DATE == 1 || PS_DATE == 2
-	float c : SV_Target;
-#else
-	float4 c0 : SV_Target0;
-	#undef NUM_RTS
-	#define NUM_RTS 1
-#if !PS_NO_COLOR1
-	float4 c1 : SV_Target1;
+
+#if PS_RETURN_COLOR
+	#if PS_DATE == 1 || PS_DATE == 2
+		float c : SV_Target;
+	#else
+		
+		float4 c0 : SV_Target0;
+
+		#undef NUM_RTS
+		#define NUM_RTS 1
+		
+		#if !PS_NO_COLOR1
+			float4 c1 : SV_Target1;
+		#endif
+	#endif
 #endif
-#endif
-#endif
-#if ZWRITE
+
+#if PS_RETURN_DEPTH
 	// In DX12 we do depth feedback loops with a color copy.
 	#if SW_DEPTH && PS_NO_COLOR1 && DX12
 		#if NUM_RTS > 0
@@ -208,16 +227,45 @@ struct PS_OUTPUT
 		float depth : SV_Depth;
 	#endif
 #endif
+
 #undef NUM_RTS
 };
 
+struct PS_OUTPUT
+{
+#if !PS_NO_COLOR
+	#if PS_DATE == 1 || PS_DATE == 2
+		float c;
+	#else
+		float4 c0;
+		#if !PS_NO_COLOR1
+			float4 c1;
+		#endif
+	#endif
+#endif
+};
+
 Texture2D<float4> Texture : register(t0);
 Texture2D<float4> Palette : register(t1);
+#if !PS_ROV_COLOR
 Texture2D<float4> RtTexture : register(t2);
+#endif
 Texture2D<float> PrimMinTexture : register(t3);
+#if !PS_ROV_DEPTH
 Texture2D<float> DepthTexture : register(t4);
+#endif
 SamplerState TextureSampler : register(s0);
 
+#if PS_ROV_COLOR
+RasterizerOrderedTexture2D<unorm float4> RtTextureRov : register(u0);
+static float4 rov_rt_value;
+#endif
+
+#if PS_ROV_DEPTH
+RasterizerOrderedTexture2D<float> DepthTextureRov : register(u1);
+static float rov_depth_value;
+#endif
+
 #ifdef DX12
 cbuffer cb1 : register(b1)
 #else
@@ -244,6 +292,38 @@ cbuffer cb1
 	float RcpScaleFactor;
 };
 
+float4 RtLoad(int2 xy)
+{
+#if PS_ROV_COLOR
+	return rov_rt_value;
+#else
+	return RtTexture.Load(int3(int2(xy), 0));
+#endif
+}
+
+float DepthLoad(int2 xy)
+{
+#if PS_ROV_DEPTH
+	return rov_depth_value;
+#else
+	return DepthTexture.Load(int3(int2(xy), 0));
+#endif
+}
+
+void RtWrite(int2 xy, float4 c)
+{
+#if PS_ROV_COLOR
+	RtTextureRov[xy] = c;
+#endif
+}
+
+void DepthWrite(int2 xy, float d)
+{
+#if PS_ROV_DEPTH
+	DepthTextureRov[xy] = d;
+#endif
+}
+
 #if (PS_AUTOMATIC_LOD != 1) && (PS_MANUAL_LOD == 1)
 float manual_lod(float uv_w)
 {
@@ -407,7 +487,7 @@ float4 sample_c_af(float2 uv, float uv_w)
 float4 sample_c(float2 uv, float uv_w, int2 xy)
 {
 #if PS_TEX_IS_FB == 1
-	return RtTexture.Load(int3(int2(xy), 0));
+	return RtLoad(xy);
 #elif PS_REGION_RECT == 1
 	return Texture.Load(int3(int2(uv), 0));
 #else
@@ -602,7 +682,7 @@ float4x4 sample_4p(uint4 u)
 uint fetch_raw_depth(int2 xy)
 {
 #if PS_TEX_IS_FB == 1
-	float4 col = RtTexture.Load(int3(xy, 0));
+	float4 col = RtLoad(xy);
 #else
 	float4 col = Texture.Load(int3(xy, 0));
 #endif
@@ -612,7 +692,7 @@ uint fetch_raw_depth(int2 xy)
 float4 fetch_raw_color(int2 xy)
 {
 #if PS_TEX_IS_FB == 1
-	return RtTexture.Load(int3(xy, 0));
+	return RtLoad(xy);
 #else
 	return Texture.Load(int3(xy, 0));
 #endif
@@ -621,7 +701,7 @@ float4 fetch_raw_color(int2 xy)
 float4 fetch_c(int2 uv)
 {
 #if PS_TEX_IS_FB == 1
-	return RtTexture.Load(int3(uv, 0));
+	return RtLoad(uv);
 #else
 	return Texture.Load(int3(uv, 0));
 #endif
@@ -1027,7 +1107,7 @@ void ps_fbmask(inout float4 C, float2 pos_xy)
 	if (PS_FBMASK)
 	{
 		float multi = PS_COLCLIP_HW ? 65535.0f : 255.0f;
-		float4 RT = trunc(RtTexture.Load(int3(pos_xy, 0)) * multi + 0.1f);
+		float4 RT = trunc(RtLoad(int2(pos_xy)) * multi + 0.1f);
 		C = (float4)(((uint4)C & ~FbMask) | ((uint4)RT & FbMask));
 	}
 }
@@ -1103,7 +1183,7 @@ void ps_blend(inout float4 Color, inout float4 As_rgba, float2 pos_xy)
 			As_rgba.rgb = (float3)1.0f;
 		}
 
-		float4 RT = SW_BLEND_NEEDS_RT ? RtTexture.Load(int3(pos_xy, 0)) : (float4)0.0f;
+		float4 RT = SW_BLEND_NEEDS_RT ? RtLoad(int2(pos_xy)) : (float4)0.0f;
 
 		if (PS_SHUFFLE && SW_BLEND_NEEDS_RT)
 		{
@@ -1236,20 +1316,49 @@ void ps_blend(inout float4 Color, inout float4 As_rgba, float2 pos_xy)
 	}
 }
 
-PS_OUTPUT ps_main(PS_INPUT input)
+#if PS_ROV_EARLYDEPTHSTENCIL
+[earlydepthstencil]
+#endif
+
+#if PS_ROV_COLOR || PS_ROV_DEPTH
+#define DISCARD rov_discard = true
+#else
+#define DISCARD discard
+#endif
+
+#if (PS_RETURN_COLOR || PS_RETURN_DEPTH)
+PS_OUTPUT_REAL ps_main(PS_INPUT input)
+#else
+void ps_main(PS_INPUT input)
+#endif
 {
 	// Must floor before depth testing.
 #if PS_ZFLOOR
 	input.p.z = floor(input.p.z * exp2(32.0f)) * exp2(-32.0f);
 #endif
 
-#if PS_ZTST == ZTST_GEQUAL
-	if (input.p.z < DepthTexture.Load(int3(input.p.xy, 0)).r)
-		discard;
-#elif PS_ZTST == ZTST_GREATER
-	if (input.p.z <= DepthTexture.Load(int3(input.p.xy, 0)).r)
-		discard;
+#if PS_ROV_COLOR
+	rov_rt_value = RtTextureRov[input.p.xy];
 #endif
+
+#if PS_ROV_DEPTH
+	rov_depth_value = DepthTextureRov[input.p.xy];
+#endif
+
+#if PS_ROV_COLOR || PS_ROV_DEPTH
+	bool rov_discard = false;
+#endif
+
+	// Use ROV discard macro for since we cannot do
+	// conditional discard based on value read from ROV.
+#if PS_ZTST == ZTST_GEQUAL
+	if (input.p.z < DepthLoad(input.p.xy))
+		DISCARD;
+#elif PS_ZTST == ZTST_GREATER
+	if (input.p.z <= DepthLoad(input.p.xy))
+		DISCARD;
+#endif
+
 	float4 C = ps_color(input);
 
 #if PS_AA1
@@ -1267,13 +1376,11 @@ PS_OUTPUT ps_main(PS_INPUT input)
 
 	bool atst_pass = atst(C);
 
-#if PS_AFAIL == AFAIL_KEEP
+#if PS_ATST != PS_ATST_NONE && PS_AFAIL == AFAIL_KEEP
 	if (!atst_pass)
 		discard;
 #endif
 
-	PS_OUTPUT output;
-
 	if (PS_SCANMSK & 2)
 	{
 		// fail depth test on prohibited lines
@@ -1284,7 +1391,7 @@ PS_OUTPUT ps_main(PS_INPUT input)
 	float4 alpha_blend = (float4)0.0f;
 	if (SW_AD_TO_HW)
 	{
-		float4 RT = PS_RTA_CORRECTION ? trunc(RtTexture.Load(int3(input.p.xy, 0)) * 128.0f + 0.1f) : trunc(RtTexture.Load(int3(input.p.xy, 0)) * 255.0f + 0.1f);
+		float4 RT = PS_RTA_CORRECTION ? trunc(RtLoad(input.p.xy) * 128.0f + 0.1f) : trunc(RtLoad(input.p.xy) * 255.0f + 0.1f);
 		alpha_blend = (float4)(RT.a / 128.0f);
 	}
 	else
@@ -1308,9 +1415,9 @@ PS_OUTPUT ps_main(PS_INPUT input)
 
 #if PS_WRITE_RG == 1
 	// Pseudo 16 bits access.
-	float rt_a = RtTexture.Load(int3(input.p.xy, 0)).g;
+	float rt_a = RtLoad(input.p.xy).g;
 #else
-	float rt_a = RtTexture.Load(int3(input.p.xy, 0)).a;
+	float rt_a = RtLoad(input.p.xy).a;
 #endif
 
 #if (PS_DATE & 3) == 1
@@ -1329,9 +1436,8 @@ PS_OUTPUT ps_main(PS_INPUT input)
 	#endif
 #endif
 
-	if (bad)
-		discard;
-
+if (bad)
+	discard;
 #endif
 
 #if PS_DATE == 3
@@ -1342,6 +1448,8 @@ PS_OUTPUT ps_main(PS_INPUT input)
 		discard;
 #endif
 
+	PS_OUTPUT output;
+
 	// Get first primitive that will write a failling alpha value
 #if PS_DATE == 1
 	// DATM == 0
@@ -1427,32 +1535,32 @@ PS_OUTPUT ps_main(PS_INPUT input)
 	alpha_blend.a = float(atst_pass);
 #endif
 
+	// Output color scaling
 #if !PS_NO_COLOR
 	output.c0.a = PS_RTA_CORRECTION ? C.a / 128.0f : C.a / 255.0f;
 	output.c0.rgb = PS_COLCLIP_HW ? float3(C.rgb / 65535.0f) : C.rgb / 255.0f;
 #if !PS_NO_COLOR1
 	output.c1 = alpha_blend;
 #endif
+#endif // !PS_NO_COLOR
 
 	// Alpha test with feedback
 #if PS_AFAIL == AFAIL_FB_ONLY
 	if (!atst_pass)
-		input.p.z = DepthTexture.Load(int3(input.p.xy, 0)).r;
+		input.p.z = DepthLoad(input.p.xy);
 #elif PS_AFAIL == AFAIL_ZB_ONLY
 	if (!atst_pass)
-		output.c0 = RtTexture.Load(int3(input.p.xy, 0));
+		output.c0 = RtLoad(input.p.xy);
 #elif PS_AFAIL == AFAIL_RGB_ONLY || PS_AFAIL == AFAIL_RGB_ONLY_SW_Z
 	if (!atst_pass)
 	{
-	output.c0.a = RtTexture.Load(int3(input.p.xy, 0)).a;
+		output.c0.a = RtLoad(input.p.xy).a;
 	#if PS_AFAIL == AFAIL_RGB_ONLY_SW_Z
-		input.p.z = DepthTexture.Load(int3(input.p.xy, 0)).r; 
+		input.p.z = DepthLoad(input.p.xy); 
 	#endif
 	}
 #endif
 
-#endif // !PS_NO_COLOR
-
 #endif // PS_DATE != 1/2
 
 #if PS_ZCLAMP
@@ -1461,18 +1569,47 @@ PS_OUTPUT ps_main(PS_INPUT input)
 
 #if PS_AA1 == PS_AA1_TRIANGLE_SW_Z
 	if (!bool(input.interior))
-		input.p.z = DepthTexture.Load(int3(input.p.xy, 0)).r; // No depth update for triangle edges.
+		input.p.z = DepthLoad(input.p.xy); // No depth update for triangle edges.
 #endif
 
-#if ZWRITE
-#if SW_DEPTH && PS_NO_COLOR1 && DX12
-	// Output color clone for feedback as well as real depth.
-	output.depth_color = input.p.z;
-#endif
-	output.depth = input.p.z;
+#if (PS_RETURN_COLOR || PS_RETURN_DEPTH)
+	// Output struct with the actual system values output semantics.
+	PS_OUTPUT_REAL output_real;
 #endif
 
-	return output;
+	// Color write back
+#if PS_RETURN_COLOR
+	#if PS_DATE == 1 || PS_DATE == 2
+		output_real.c = output.c;
+	#else
+		output_real.c0 = output.c0;
+		#if !PS_NO_COLOR1
+			output_real.c1 = output.c1;
+		#endif
+	#endif
+#elif PS_RETURN_COLOR_ROV
+	output.c0 = (rov_discard | (FbMask == 0xFFu)) ? RtLoad(input.p.xy) : output.c0;
+
+	RtWrite(input.p.xy, output.c0);
+#endif
+
+	// Depth write back
+#if PS_RETURN_DEPTH
+	output_real.depth = input.p.z;
+	#if SW_DEPTH && PS_NO_COLOR1 && DX12
+		// Output color clone for feedback.
+		output_real.depth_color = input.p.z;
+	#endif
+#elif PS_RETURN_DEPTH_ROV
+	#if SW_DEPTH
+		input.p.z = rov_discard ? DepthLoad(input.p.xy) : input.p.z;
+	#endif
+	DepthWrite(input.p.xy, input.p.z);
+#endif
+
+#if (PS_RETURN_COLOR || PS_RETURN_DEPTH)
+	return output_real;
+#endif
 }
 
 #endif // PIXEL_SHADER

bin/resources/shaders/opengl/tfx_fs.glsl

@@ -40,6 +40,12 @@
 #define PS_AA1_TRIANGLE_SW_Z 3
 #endif
 
+#ifndef PS_ROV_DEPTH_NONE
+#define PS_ROV_DEPTH_NONE 0
+#define PS_ROV_DEPTH_READ_WRITE 1
+#define PS_ROV_DEPTH_READ_ONLY 2
+#endif
+
 // TEX_COORD_DEBUG output the uv coordinate as color. It is useful
 // to detect bad sampling due to upscaling
 //#define TEX_COORD_DEBUG
@@ -59,11 +65,17 @@
 #define NEEDS_DEPTH_FOR_ZTST (PS_ZTST == ZTST_GEQUAL || PS_ZTST == ZTST_GREATER)
 #define NEEDS_DEPTH_FOR_AA1 (PS_AA1 == PS_AA1_TRIANGLE_SW_Z)
 
-#define NEEDS_RT (NEEDS_RT_EARLY || NEEDS_RT_FOR_AFAIL || (!PS_PRIMID_INIT && (PS_FBMASK || SW_BLEND_NEEDS_RT || SW_AD_TO_HW)) || PS_COLOR_FEEDBACK)
+#define NEEDS_RT (NEEDS_RT_EARLY || NEEDS_RT_FOR_AFAIL || (!PS_PRIMID_INIT && (PS_FBMASK || SW_BLEND_NEEDS_RT || SW_AD_TO_HW)))
 #define NEEDS_TEX (PS_TFX != 4)
 #define SW_DEPTH (NEEDS_DEPTH_FOR_AFAIL || NEEDS_DEPTH_FOR_ZTST || NEEDS_DEPTH_FOR_AA1)
 #define ZWRITE (SW_DEPTH || PS_ZCLAMP || PS_ZFLOOR)
 
+#define PS_RETURN_COLOR_ROV (!PS_NO_COLOR && PS_ROV_COLOR)
+#define PS_RETURN_COLOR (!PS_NO_COLOR && !PS_ROV_COLOR)
+#define PS_RETURN_DEPTH_ROV (PS_ROV_DEPTH == PS_ROV_DEPTH_READ_WRITE)
+#define PS_RETURN_DEPTH (ZWRITE && !PS_ROV_DEPTH)
+#define PS_ROV_EARLYDEPTHSTENCIL (PS_ROV_COLOR && !PS_ROV_DEPTH && !ZWRITE)
+
 layout(std140, binding = 0) uniform cb21
 {
 	vec3 FogColor;
@@ -121,28 +133,42 @@ in SHADER
 	#if defined(GL_EXT_shader_framebuffer_fetch)
 		#undef TARGET_0_QUALIFIER
 		#define TARGET_0_QUALIFIER inout
-		#define LAST_FRAG_COLOR SV_Target0
+		#define LAST_FRAG_COLOR o_col0
 	#elif defined(GL_ARM_shader_framebuffer_fetch)
 		#define LAST_FRAG_COLOR gl_LastFragColorARM
 	#endif
 #endif
 
-#if !PS_NO_COLOR && !PS_NO_COLOR1
-	// Same buffer but 2 colors for dual source blending
-	layout(location = 0, index = 0) TARGET_0_QUALIFIER vec4 SV_Target0;
-	layout(location = 0, index = 1) out vec4 SV_Target1;
-#elif !PS_NO_COLOR
-	layout(location = 0) TARGET_0_QUALIFIER vec4 SV_Target0;
+#if PS_RETURN_COLOR
+	#if !PS_NO_COLOR && !PS_NO_COLOR1
+		// Same buffer but 2 colors for dual source blending
+		layout(location = 0, index = 0) TARGET_0_QUALIFIER vec4 o_col0;
+		layout(location = 0, index = 1) out vec4 o_col1;
+	#elif !PS_NO_COLOR
+		layout(location = 0) TARGET_0_QUALIFIER vec4 o_col0;
+	#endif
+#elif PS_RETURN_COLOR_ROV
+	vec4 o_col0;
 #endif
 
 // Depth feedback mode 2 is for depth as color.
 // Use FB fetch for the feedback if it's available.
 #if SW_DEPTH && PS_NO_COLOR1 && (DEPTH_FEEDBACK_SUPPORT == 2)
-#if HAS_FRAMEBUFFER_FETCH
-	layout(location = 1) inout float SV_Target1;
-#else
-	layout(location = 1) out float SV_Target1;
+	#if HAS_FRAMEBUFFER_FETCH
+		layout(location = 1) inout float o_col1;
+	#else
+		layout(location = 1) out float o_col1;
+	#endif
 #endif
+
+#if PS_ROV_COLOR
+	layout(binding = 0, rgba8) uniform restrict coherent image2D RtImageRov;
+	vec4 rov_rt_value = vec4(0.0f, 0.0f, 0.0f, 1.0f);
+#endif
+
+#if PS_ROV_DEPTH
+	layout(binding = 1, r32f) uniform restrict coherent image2D DepthImageRov;
+	float rov_depth_value = 0.0f;
 #endif
 
 #if NEEDS_TEX
@@ -161,17 +187,19 @@ layout(binding = 3) uniform sampler2D img_prim_min;
 // Depth feedback mode 1 binds depth buffer directly as a texture.
 // Depth feedback mode 2 (depth as color) can use FB fetch for the feedback,
 // in which case we don't need to explicitly bind depth as a texture.
-#if (DEPTH_FEEDBACK_SUPPORT == 1 || (DEPTH_FEEDBACK_SUPPORT == 2 && !HAS_FRAMEBUFFER_FETCH)) && SW_DEPTH
+#if (DEPTH_FEEDBACK_SUPPORT == 1 || (DEPTH_FEEDBACK_SUPPORT == 2 && !HAS_FRAMEBUFFER_FETCH)) && SW_DEPTH && !PS_ROV_DEPTH
 layout(binding = 4) uniform sampler2D DepthSampler;
 #endif
 
-#if ZWRITE && PS_HAS_CONSERVATIVE_DEPTH && !SW_DEPTH
+#if ZWRITE && PS_HAS_CONSERVATIVE_DEPTH && !SW_DEPTH && !PS_ROV_DEPTH
 layout(depth_less) out float gl_FragDepth;
 #endif
 
 vec4 sample_from_rt()
 {
-#if !NEEDS_RT
+#if PS_ROV_COLOR
+	return rov_rt_value;
+#elif !NEEDS_RT
 	return vec4(0.0);
 #elif HAS_FRAMEBUFFER_FETCH
 	return LAST_FRAG_COLOR;
@@ -180,14 +208,16 @@ vec4 sample_from_rt()
 #endif
 }
 
-vec4 sample_from_depth()
+float sample_from_depth()
 {
-#if !SW_DEPTH
-	return vec4(0.0);
+#if PS_ROV_DEPTH
+	return rov_depth_value;
+#elif !SW_DEPTH
+	return 0.0f;
 #elif HAS_FRAMEBUFFER_FETCH && (DEPTH_FEEDBACK_SUPPORT == 2)
-	return SV_Target1;
+	return o_col1;
 #else
-	return texelFetch(DepthSampler, ivec2(gl_FragCoord.xy), 0);
+	return texelFetch(DepthSampler, ivec2(gl_FragCoord.xy), 0).r;
 #endif
 }
 
@@ -1183,6 +1213,20 @@ float As = As_rgba.a;
 #endif
 }
 
+#if (PS_ROV_COLOR || PS_ROV_DEPTH) && (USE_ARB_FSI || USE_NV_FSI)
+layout(pixel_interlock_ordered) in;
+#endif
+
+#if PS_ROV_EARLYDEPTHSTENCIL
+layout(early_fragment_tests) in;
+#endif
+
+#if PS_ROV_COLOR || PS_ROV_DEPTH
+#define DISCARD rov_discard = true
+#else
+#define DISCARD discard
+#endif
+
 void ps_main()
 {
 	float input_z = gl_FragCoord.z;
@@ -1192,18 +1236,40 @@ void ps_main()
 	input_z = floor(input_z * exp2(32.0f)) * exp2(-32.0f);
 #endif
 
+#if PS_ROV_COLOR || PS_ROV_DEPTH
+	#if USE_ARB_FSI
+		beginInvocationInterlockARB();
+	#elif USE_NV_FSI
+		beginInvocationInterlockNV();
+	#elif USE_INTEL_FSI
+		beginFragmentShaderOrderingINTEL();
+	#endif
+#endif
+
+#if PS_ROV_COLOR
+	rov_rt_value = imageLoad(RtImageRov, ivec2(gl_FragCoord.xy));
+#endif
+
+#if PS_ROV_DEPTH
+	rov_depth_value = imageLoad(DepthImageRov, ivec2(gl_FragCoord.xy)).r;
+#endif
+
+#if PS_ROV_COLOR || PS_ROV_DEPTH
+	bool rov_discard = false;
+#endif
+
 #if PS_ZTST == ZTST_GEQUAL
-	if (input_z < sample_from_depth().r)
-		discard;
+	if (input_z < sample_from_depth())
+		DISCARD;
 #elif PS_ZTST == ZTST_GREATER
-	if (input_z <= sample_from_depth().r)
-		discard;
+	if (input_z <= sample_from_depth())
+		DISCARD;
 #endif
 
 #if PS_SCANMSK & 2
 	// fail depth test on prohibited lines
 	if ((int(gl_FragCoord.y) & 1) == (PS_SCANMSK & 1))
-		discard;
+		DISCARD;
 #endif
 
 #if PS_DATE >= 5
@@ -1232,7 +1298,7 @@ void ps_main()
 #endif
 
 	if (bad) {
-		discard;
+		DISCARD;
 	}
 
 #endif
@@ -1243,7 +1309,7 @@ void ps_main()
 	// the bad alpha value so we must keep it.
 
 	if (gl_PrimitiveID > stencil_ceil) {
-		discard;
+		DISCARD;
 	}
 #endif
 
@@ -1264,9 +1330,9 @@ void ps_main()
 
 	bool atst_pass = atst(C);
 
-#if PS_AFAIL == AFAIL_KEEP
+#if PS_ATST != PS_ATST_NONE && PS_AFAIL == AFAIL_KEEP
 	if (!atst_pass)
-		discard;
+		DISCARD;
 #endif
 
 #if SW_AD_TO_HW
@@ -1293,12 +1359,12 @@ void ps_main()
 #if PS_DATE == 1
 	// DATM == 0
 	// Pixel with alpha equal to 1 will failed (128-255)
-	SV_Target0 = (C.a > 127.5f) ? vec4(gl_PrimitiveID) : vec4(0x7FFFFFFF);
+	o_col0 = (C.a > 127.5f) ? vec4(gl_PrimitiveID) : vec4(0x7FFFFFFF);
 	return;
 #elif PS_DATE == 2
 	// DATM == 1
 	// Pixel with alpha equal to 0 will failed (0-127)
-	SV_Target0 = (C.a < 127.5f) ? vec4(gl_PrimitiveID) : vec4(0x7FFFFFFF);
+	o_col0 = (C.a < 127.5f) ? vec4(gl_PrimitiveID) : vec4(0x7FFFFFFF);
 	return;
 #endif
 
@@ -1372,7 +1438,7 @@ void ps_main()
 	// Alpha test with feedback
 	#if PS_AFAIL == AFAIL_FB_ONLY
 		if (!atst_pass)
-			input_z = sample_from_depth().r;
+			input_z = sample_from_depth();
 	#elif PS_AFAIL == AFAIL_ZB_ONLY
 		if (!atst_pass)
 			C = sample_from_rt();
@@ -1381,17 +1447,17 @@ void ps_main()
 		{
 			C.a = sample_from_rt().a;
 		#if PS_AFAIL == AFAIL_RGB_ONLY_SW_Z
-			input_z = sample_from_depth().r;
+			input_z = sample_from_depth();
 		#endif
 		}
 	#endif
 
-	// Warning: do not write SV_Target0 until the end since the value might be needed for
+	// Warning: do not write o_col0 until the end since the value might be needed for
 	// FB fetch in sample_from_rt().
-	SV_Target0 = C;
+	o_col0 = C;
 
 	#if !PS_NO_COLOR1
-		SV_Target1 = alpha_blend;
+		o_col1 = alpha_blend;
 	#endif
 #endif
 
@@ -1401,18 +1467,40 @@ void ps_main()
 
 #if PS_AA1 == PS_AA1_TRIANGLE_SW_Z
 	if (!bool(PSin.interior))
-		input_z = sample_from_depth().r; // No depth update for triangle edges.
+		input_z = sample_from_depth(); // No depth update for triangle edges.
 #endif
 
-#if ZWRITE
+// Writing back color (result already written to o_col0 for non-ROV)
+#if PS_RETURN_COLOR_ROV
+	bvec4 discard_channels = bvec4(uvec4(rov_discard) | uvec4(equal(FbMask, uvec4(0xFFu))));
+	o_col0 = mix(o_col0, sample_from_rt(), discard_channels);
+
+	imageStore(RtImageRov, ivec2(gl_FragCoord.xy), o_col0);
+#endif
+
+// Writing back depth
+#if PS_RETURN_DEPTH
 	#if SW_DEPTH && PS_NO_COLOR1 && (DEPTH_FEEDBACK_SUPPORT == 2)
 		// Depth as color write. For depth as color feedback we write to both
 		// color copy and real depth to avoid having to copy back to real depth.
-		// Warning: do not write SV_Target1 until the end since the value might
+		// Warning: do not write o_col1 until the end since the value might
 		// be needed for FB fetch in sample_from_depth().
-		SV_Target1 = input_z;
+		o_col1 = input_z;
 	#endif
 	// Standard depth write.
 	gl_FragDepth = input_z;
+#elif PS_RETURN_DEPTH_ROV
+	input_z = rov_discard ? sample_from_depth() : input_z;
+
+	imageStore(DepthImageRov, ivec2(gl_FragCoord.xy), vec4(input_z, 0, 0, 1.0f));
+#endif
+
+#if (PS_ROV_COLOR || PS_ROV_DEPTH)
+	#if USE_ARB_FSI
+		endInvocationInterlockARB();
+	#elif USE_NV_FSI
+		endInvocationInterlockNV();
+	#endif
+	// No end invocation for Intel fragment shader ordering.
 #endif
 }

bin/resources/shaders/vulkan/tfx.glsl

@@ -522,6 +522,12 @@ void main()
 #define PS_AA1_TRIANGLE_SW_Z 3
 #endif
 
+#ifndef PS_ROV_DEPTH_NONE
+#define PS_ROV_DEPTH_NONE 0
+#define PS_ROV_DEPTH_READ_WRITE 1
+#define PS_ROV_DEPTH_READ_ONLY 2
+#endif
+
 #ifndef PS_FST
 #define PS_FST 0
 #define PS_WMS 0
@@ -574,6 +580,8 @@ void main()
 #define PS_NO_COLOR1 0
 #define PS_DATE 0
 #define PS_TEX_IS_FB 0
+#define PS_ROV_COLOR 0
+#define PS_ROV_DEPTH 0
 #endif
 
 #define SW_BLEND (PS_BLEND_A || PS_BLEND_B || PS_BLEND_D)
@@ -586,7 +594,13 @@ void main()
 
 #define PS_FEEDBACK_LOOP_IS_NEEDED_RT (PS_TEX_IS_FB == 1 || AFAIL_NEEDS_RT || PS_FBMASK || SW_BLEND_NEEDS_RT || SW_AD_TO_HW || (PS_DATE >= 5))
 #define PS_FEEDBACK_LOOP_IS_NEEDED_DEPTH (AFAIL_NEEDS_DEPTH || ZTST_NEEDS_DEPTH || AA1_NEEDS_DEPTH)
-#define ZWRITE (PS_ZCLAMP || PS_ZFLOOR || PS_FEEDBACK_LOOP_IS_NEEDED_DEPTH)
+#define ZWRITE (PS_ZCLAMP || PS_ZFLOOR || SW_DEPTH || PS_FEEDBACK_LOOP_IS_NEEDED_DEPTH)
+
+#define PS_RETURN_COLOR_ROV (!PS_NO_COLOR && PS_ROV_COLOR)
+#define PS_RETURN_COLOR (!PS_NO_COLOR && !PS_ROV_COLOR)
+#define PS_RETURN_DEPTH_ROV (PS_ROV_DEPTH == PS_ROV_DEPTH_READ_WRITE)
+#define PS_RETURN_DEPTH (ZWRITE && !PS_ROV_DEPTH)
+#define PS_ROV_EARLYDEPTHSTENCIL (PS_ROV_COLOR && !PS_ROV_DEPTH && !ZWRITE)
 
 #define NEEDS_TEX (PS_TFX != 4)
 
@@ -625,11 +639,27 @@ layout(location = 0) in VSOutput
 	flat uint interior; // 1 for triangle interior; 0 for edge;
 } vsIn;
 
-#if !PS_NO_COLOR && !PS_NO_COLOR1
-layout(location = 0, index = 0) out vec4 o_col0;
-layout(location = 0, index = 1) out vec4 o_col1;
-#elif !PS_NO_COLOR
-layout(location = 0) out vec4 o_col0;
+#if PS_RETURN_COLOR
+	#if !PS_NO_COLOR1
+		layout(location = 0, index = 0) out vec4 o_col0;
+		layout(location = 0, index = 1) out vec4 o_col1;
+	#elif !PS_NO_COLOR
+		layout(location = 0) out vec4 o_col0;
+	#endif
+#elif PS_RETURN_COLOR_ROV
+	vec4 o_col0;
+#endif
+
+#if PS_ROV_COLOR
+	layout(set = 1, binding = 5, rgba8) uniform restrict coherent image2D RtImageRov;
+	vec4 rov_rt_value;
+	vec4 sample_from_rt() { return rov_rt_value; }
+#endif
+
+#if PS_ROV_DEPTH
+	layout(set = 1, binding = 6, r32f) uniform restrict coherent image2D DepthImageRov;
+	float rov_depth_value;
+	float sample_from_depth() { return rov_depth_value; }
 #endif
 
 #if NEEDS_TEX
@@ -639,27 +669,27 @@ layout(set = 1, binding = 1) uniform texture2D Palette;
 
 #if PS_FEEDBACK_LOOP_IS_NEEDED_RT || PS_FEEDBACK_LOOP_IS_NEEDED_DEPTH
 	#if defined(DISABLE_TEXTURE_BARRIER) || defined(HAS_FEEDBACK_LOOP_LAYOUT)
-		#if PS_FEEDBACK_LOOP_IS_NEEDED_RT
+		#if (PS_FEEDBACK_LOOP_IS_NEEDED_RT && !PS_ROV_COLOR)
 			layout(set = 1, binding = 2) uniform texture2D RtSampler;
 			vec4 sample_from_rt() { return texelFetch(RtSampler, ivec2(gl_FragCoord.xy), 0); }
 		#endif
-		#if PS_FEEDBACK_LOOP_IS_NEEDED_DEPTH
+		#if (PS_FEEDBACK_LOOP_IS_NEEDED_DEPTH && !PS_ROV_DEPTH)
 			layout(set = 1, binding = 4) uniform texture2D DepthSampler;
-			vec4 sample_from_depth() { return texelFetch(DepthSampler, ivec2(gl_FragCoord.xy), 0); }
+			float sample_from_depth() { return texelFetch(DepthSampler, ivec2(gl_FragCoord.xy), 0).r; }
 		#endif
 	#else
 		// Must consider each case separately since the input attachment indices must be consecutive.
-		#if PS_FEEDBACK_LOOP_IS_NEEDED_RT && PS_FEEDBACK_LOOP_IS_NEEDED_DEPTH
+		#if (PS_FEEDBACK_LOOP_IS_NEEDED_RT && !PS_ROV_COLOR) && (PS_FEEDBACK_LOOP_IS_NEEDED_DEPTH && !PS_ROV_DEPTH)
 			layout(input_attachment_index = 0, set = 1, binding = 2) uniform subpassInput RtSampler;
 			layout(input_attachment_index = 1, set = 1, binding = 4) uniform subpassInput DepthSampler;
 			vec4 sample_from_rt() { return subpassLoad(RtSampler); }
-			vec4 sample_from_depth() { return subpassLoad(DepthSampler); }
-		#elif PS_FEEDBACK_LOOP_IS_NEEDED_RT
+			float sample_from_depth() { return subpassLoad(DepthSampler).r; }
+		#elif (PS_FEEDBACK_LOOP_IS_NEEDED_RT && !PS_ROV_COLOR)
 			layout(input_attachment_index = 0, set = 1, binding = 2) uniform subpassInput RtSampler;
 			vec4 sample_from_rt() { return subpassLoad(RtSampler); }
-		#elif PS_FEEDBACK_LOOP_IS_NEEDED_DEPTH
+		#elif (PS_FEEDBACK_LOOP_IS_NEEDED_DEPTH && !PS_ROV_DEPTH)
 			layout(input_attachment_index = 0, set = 1, binding = 4) uniform subpassInput DepthSampler;
-			vec4 sample_from_depth() { return subpassLoad(DepthSampler); }
+			float sample_from_depth() { return subpassLoad(DepthSampler).r; }
 		#endif
 	#endif
 #endif
@@ -1604,7 +1634,7 @@ void ps_blend(inout vec4 Color, inout vec4 As_rgba)
 		#elif PS_BLEND_MIX == 1
 			Color.rgb = ((A - B) * C_clamped + D) - (124.0f/256.0f);
 		#else
-				Color.rgb = trunc((A - B) * C + D);
+			Color.rgb = trunc((A - B) * C + D);
 		#endif
 
 		#if PS_BLEND_HW == 1
@@ -1678,6 +1708,20 @@ void ps_blend(inout vec4 Color, inout vec4 As_rgba)
 	#endif
 }
 
+#if PS_ROV_COLOR || PS_ROV_DEPTH
+layout(pixel_interlock_ordered) in;
+#endif
+
+#if PS_ROV_EARLYDEPTHSTENCIL
+layout(early_fragment_tests) in;
+#endif
+
+#if PS_ROV_COLOR || PS_ROV_DEPTH
+#define DISCARD rov_discard = true
+#else
+#define DISCARD discard
+#endif
+
 void main()
 {
 	float input_z = gl_FragCoord.z;
@@ -1687,18 +1731,34 @@ void main()
 	input_z = floor(input_z * exp2(32.0f)) * exp2(-32.0f);
 #endif
 
+#if PS_ROV_COLOR || PS_ROV_DEPTH
+	beginInvocationInterlockARB();
+#endif
+
+#if PS_ROV_COLOR
+	rov_rt_value = imageLoad(RtImageRov, ivec2(gl_FragCoord.xy));
+#endif
+
+#if PS_ROV_DEPTH
+	rov_depth_value = imageLoad(DepthImageRov, ivec2(gl_FragCoord.xy)).r;
+#endif
+
+#if PS_ROV_COLOR || PS_ROV_DEPTH
+	bool rov_discard = gl_HelperInvocation;
+#endif
+
 #if PS_ZTST == ZTST_GEQUAL
-	if (input_z < sample_from_depth().r)
-		discard;
+	if (input_z < sample_from_depth())
+		DISCARD;
 #elif PS_ZTST == ZTST_GREATER
-	if (input_z <= sample_from_depth().r)
-		discard;
+	if (input_z <= sample_from_depth())
+		DISCARD;
 #endif
 
 #if PS_SCANMSK & 2
 	// fail depth test on prohibited lines
 	if ((int(gl_FragCoord.y) & 1) == (PS_SCANMSK & 1))
-		discard;
+		DISCARD;
 #endif
 #if PS_DATE >= 5
 
@@ -1726,7 +1786,7 @@ void main()
 #endif
 
 	if (bad) {
-		discard;
+		DISCARD;
 	}
 
 #endif		// PS_DATE >= 5
@@ -1737,7 +1797,7 @@ void main()
 	// the bad alpha value so we must keep it.
 
 	if (gl_PrimitiveID > stencil_ceil) {
-		discard;
+		DISCARD;
 	}
 #endif
 
@@ -1758,9 +1818,10 @@ void main()
 
 	bool atst_pass = atst(C);
 
-#if PS_AFAIL == AFAIL_KEEP
-	if (!atst_pass)
-		discard;
+#if PS_ATST != PS_ATST_NONE && PS_AFAIL == AFAIL_KEEP
+	if (!atst_pass) {
+		DISCARD;
+	}
 #endif
 
 #if SW_AD_TO_HW
@@ -1852,6 +1913,7 @@ void main()
 		alpha_blend.a = float(atst_pass);
 	#endif
 
+	// Output color scaling
 	#if !PS_NO_COLOR
 		#if PS_RTA_CORRECTION
 			o_col0.a = C.a / 128.0f;
@@ -1870,7 +1932,7 @@ void main()
 		// Alpha test with feedback
 		#if PS_AFAIL == AFAIL_FB_ONLY
 			if (!atst_pass)
-				input_z = sample_from_depth().r;
+				input_z = sample_from_depth();
 		#elif PS_AFAIL == AFAIL_ZB_ONLY
 			if (!atst_pass)
 				o_col0 = sample_from_rt();
@@ -1879,7 +1941,7 @@ void main()
 			{
 				o_col0.a = sample_from_rt().a;
 			#if PS_AFAIL == AFAIL_RGB_ONLY_SW_Z
-				input_z = sample_from_depth().r;
+				input_z = sample_from_depth();
 			#endif
 			}
 		#endif
@@ -1891,11 +1953,28 @@ void main()
 	
 	#if PS_AA1 == PS_AA1_TRIANGLE_SW_Z
 		if (!bool(vsIn.interior))
-			input_z = sample_from_depth().r; // No depth update for triangle edges.
+			input_z = sample_from_depth(); // No depth update for triangle edges.
 	#endif
 	
-	#if ZWRITE
+	// Writing back color (result already written to o_col0 for non-ROV)
+	#if PS_RETURN_COLOR_ROV
+		bvec4 discard_channels = bvec4(uvec4(rov_discard) | uvec4(equal(FbMask, uvec4(0xFFu))));
+		o_col0 = mix(o_col0, sample_from_rt(), discard_channels);
+
+		imageStore(RtImageRov, ivec2(gl_FragCoord.xy), o_col0);
+	#endif
+	
+	// Writing back depth
+	#if PS_RETURN_DEPTH
 		gl_FragDepth = input_z;
+	#elif PS_RETURN_DEPTH_ROV
+		input_z = rov_discard ? sample_from_depth() : input_z;
+
+		imageStore(DepthImageRov, ivec2(gl_FragCoord.xy), vec4(input_z, 0, 0, 1.0f));
+	#endif
+
+	#if PS_ROV_COLOR || PS_ROV_DEPTH
+		endInvocationInterlockARB();
 	#endif
 #endif // PS_DATE
 }

pcsx2-gsrunner/Main.cpp

@@ -91,6 +91,9 @@ static double s_last_barriers = 0;
 static double s_last_copies = 0;
 static double s_last_uploads = 0;
 static double s_last_readbacks = 0;
+static double s_last_depth_copies_rov = 0;
+static double s_last_draws_rov = 0;
+static double s_last_barriers_rov = 0;
 static u64 s_total_internal_draws = 0;
 static u64 s_total_draws = 0;
 static u64 s_total_render_passes = 0;
@@ -98,6 +101,9 @@ static u64 s_total_barriers = 0;
 static u64 s_total_copies = 0;
 static u64 s_total_uploads = 0;
 static u64 s_total_readbacks = 0;
+static u64 s_total_depth_copies_rov = 0;
+static u64 s_total_draws_rov = 0;
+static u64 s_total_barriers_rov = 0;
 static u32 s_total_frames = 0;
 static u32 s_total_drawn_frames = 0;
 
@@ -287,6 +293,9 @@ void Host::BeginPresentFrame()
 		update_stat(GSPerfMon::TextureCopies, s_total_copies, s_last_copies);
 		update_stat(GSPerfMon::TextureUploads, s_total_uploads, s_last_uploads);
 		update_stat(GSPerfMon::Readbacks, s_total_readbacks, s_last_readbacks);
+		update_stat(GSPerfMon::DepthCopiesROV, s_total_depth_copies_rov, s_last_depth_copies_rov);
+		update_stat(GSPerfMon::DrawCallsROV, s_total_draws_rov, s_last_draws_rov);
+		update_stat(GSPerfMon::BarriersROV, s_total_barriers_rov, s_last_barriers_rov);
 
 		const bool idle_frame = s_total_frames && (last_draws == s_total_internal_draws && last_uploads == s_total_uploads);
 
@@ -913,6 +922,9 @@ void GSRunner::DumpStats()
 	Console.WriteLn(fmt::format("@HWSTAT@ Copies: {} (avg {})", s_total_copies, static_cast<u64>(std::ceil(s_total_copies / static_cast<double>(s_total_drawn_frames)))));
 	Console.WriteLn(fmt::format("@HWSTAT@ Uploads: {} (avg {})", s_total_uploads, static_cast<u64>(std::ceil(s_total_uploads / static_cast<double>(s_total_drawn_frames)))));
 	Console.WriteLn(fmt::format("@HWSTAT@ Readbacks: {} (avg {})", s_total_readbacks, static_cast<u64>(std::ceil(s_total_readbacks / static_cast<double>(s_total_drawn_frames)))));
+	Console.WriteLn(fmt::format("@HWSTAT@ Depth Copies (ROV): {} (avg {})", s_total_depth_copies_rov, static_cast<u64>(std::ceil(s_total_depth_copies_rov / static_cast<double>(s_total_drawn_frames)))));
+	Console.WriteLn(fmt::format("@HWSTAT@ Draws Calls (ROV): {} (avg {})", s_total_draws_rov, static_cast<u64>(std::ceil(s_total_draws_rov / static_cast<double>(s_total_drawn_frames)))));
+	Console.WriteLn(fmt::format("@HWSTAT@ Barriers (ROV): {} (avg {})", s_total_barriers_rov, static_cast<u64>(std::ceil(s_total_barriers_rov / static_cast<double>(s_total_drawn_frames)))));
 	if (s_perf_enable)
 	{
 		Console.WriteLn(fmt::format("@HWSTAT@ Minimum Frame Time: {:.3f} ms ({:.3f} FPS)", PerformanceMetrics::GetMinimumFrameTime(), 1000.0f / PerformanceMetrics::GetMinimumFrameTime()));

pcsx2-qt/Settings/GraphicsAdvancedSettingsTab.ui

@@ -126,6 +126,13 @@
           </property>
          </widget>
         </item>
+        <item row="3" column="1">
+         <widget class="QCheckBox" name="rovBarriersVK">
+          <property name="text">
+           <string>ROV Barriers Vulkan</string>
+          </property>
+         </widget>
+        </item>
        </layout>
       </item>
       <item row="6" column="0">
@@ -353,6 +360,7 @@
   <tabstop>gsDumpCompression</tabstop>
   <tabstop>texturePreloading</tabstop>
   <tabstop>exclusiveFullscreenControl</tabstop>
+  <tabstop>rovBarriersVK</tabstop>
   <tabstop>useBlitSwapChain</tabstop>
   <tabstop>extendedUpscales</tabstop>
   <tabstop>disableMailboxPresentation</tabstop>

pcsx2-qt/Settings/GraphicsHardwareRenderingSettingsTab.ui

@@ -72,6 +72,13 @@
       </widget>
      </item>
      <item row="1" column="1">
+      <widget class="QCheckBox" name="rov">
+       <property name="text">
+        <string>ROV</string>
+       </property>
+      </widget>
+     </item>
+     <item row="2" column="0">
       <widget class="QCheckBox" name="enableHWFixes">
        <property name="text">
         <string>Manual Hardware Renderer Fixes</string>
@@ -243,6 +250,7 @@
   <tabstop>mipmapping</tabstop>
   <tabstop>accurateAlphaTest</tabstop>
   <tabstop>hwAA1</tabstop>
+  <tabstop>rov</tabstop>
   <tabstop>enableHWFixes</tabstop>
  </tabstops>
  <resources/>

pcsx2-qt/Settings/GraphicsSettingsWidget.cpp

@@ -113,6 +113,7 @@ GraphicsSettingsWidget::GraphicsSettingsWidget(SettingsWindow* settings_dialog,
 	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_hw.mipmapping, "EmuCore/GS", "hw_mipmap", true);
 	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_hw.accurateAlphaTest, "EmuCore/GS", "HWAccurateAlphaTest", false);
 	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_hw.hwAA1, "EmuCore/GS", "HWAA1", false);
+	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_hw.rov, "EmuCore/GS", "HWROV", false);
 	SettingWidgetBinder::BindWidgetToIntSetting(
 		sif, m_hw.blending, "EmuCore/GS", "accurate_blending_unit", static_cast<int>(AccBlendLevel::Basic));
 	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_hw.enableHWFixes, "EmuCore/GS", "UserHacks", false);
@@ -239,6 +240,7 @@ GraphicsSettingsWidget::GraphicsSettingsWidget(SettingsWindow* settings_dialog,
 	SettingWidgetBinder::BindWidgetToFloatSetting(sif, m_advanced.palFrameRate, "EmuCore/GS", "FrameRatePAL", 50.00f);
 	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_advanced.spinCPUDuringReadbacks, "EmuCore/GS", "HWSpinCPUForReadbacks", false);
 	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_advanced.spinGPUDuringReadbacks, "EmuCore/GS", "HWSpinGPUForReadbacks", false);
+	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_advanced.rovBarriersVK, "EmuCore/GS", "HWROVBarriersVK", false);
 	SettingWidgetBinder::BindWidgetToIntSetting(sif, m_advanced.texturePreloading, "EmuCore/GS", "texture_preloading", static_cast<int>(TexturePreloadingLevel::Off));
 
 	setTabVisible(m_advanced_tab, QtHost::ShouldShowAdvancedSettings());
@@ -497,6 +499,9 @@ GraphicsSettingsWidget::GraphicsSettingsWidget(SettingsWindow* settings_dialog,
 		dialog()->registerWidgetHelp(
 			m_hw.hwAA1, tr("AA1"), tr("Unchecked"), tr("Enables AA1 (PS2 antialiasing), which some games require to render correctly. This may result in a heavy performance penalty."));
 
+		dialog()->registerWidgetHelp(
+			m_hw.rov, tr("ROV"), tr("Unchecked"), tr("Enables ROV (Rasterizer Ordered View), which allows feedback loops to be executed with fewer draw calls. Can improve performance in feedback heavy games with higher accuracy settings."));
+
 		dialog()->registerWidgetHelp(
 			m_hw.textureFiltering, tr("Texture Filtering"), tr("Bilinear (PS2)"),
 			tr("Changes what filtering algorithm is used to map textures to surfaces.<br> "

pcsx2/Config.h

@@ -793,6 +793,9 @@ struct Pcsx2Config
 					HWMipmap : 1,
 					HWAccurateAlphaTest : 1,
 					HWAA1 : 1,
+					HWROV : 1,
+					HWROVLogging : 1,
+					HWROVBarriersVK : 1,
 					ManualUserHacks : 1,
 					UserHacks_AlignSpriteX : 1,
 					UserHacks_CPUFBConversion : 1,

pcsx2/GS/GS.cpp

@@ -701,16 +701,36 @@ void GSgetStats(SmallStringBase& info)
 	}
 	else
 	{
-		info.format("{} HW | {} PRIM | {} DRW | {} DRWC | {} BAR | {} RP | {} RB | {} TC | {} TU",
-			api_name,
-			(int)pm.Get(GSPerfMon::Prim),
-			(int)pm.Get(GSPerfMon::Draw),
-			(int)std::ceil(pm.Get(GSPerfMon::DrawCalls)),
-			(int)std::ceil(pm.Get(GSPerfMon::Barriers)),
-			(int)std::ceil(pm.Get(GSPerfMon::RenderPasses)),
-			(int)std::ceil(pm.Get(GSPerfMon::Readbacks)),
-			(int)std::ceil(pm.Get(GSPerfMon::TextureCopies)),
-			(int)std::ceil(pm.Get(GSPerfMon::TextureUploads)));
+		if (!GSConfig.HWROV)
+		{
+			info.format("{} HW | {} PRIM | {} DRW | {} DRWC | {} BAR | {} RP | {} RB | {} TC | {} TU",
+				api_name,
+				(int)pm.Get(GSPerfMon::Prim),
+				(int)pm.Get(GSPerfMon::Draw),
+				(int)std::ceil(pm.Get(GSPerfMon::DrawCalls)),
+				(int)std::ceil(pm.Get(GSPerfMon::Barriers)),
+				(int)std::ceil(pm.Get(GSPerfMon::RenderPasses)),
+				(int)std::ceil(pm.Get(GSPerfMon::Readbacks)),
+				(int)std::ceil(pm.Get(GSPerfMon::TextureCopies)),
+				(int)std::ceil(pm.Get(GSPerfMon::TextureUploads)));
+		}
+		else
+		{
+			// Add ROV stats along standard stats.
+			info.format("{} HW | {} PRIM | {} DRW | {}/{} DRWC | {}/{} BAR | {} RP | {} RB | {}/{} TC | {} TU",
+				api_name,
+				(int)pm.Get(GSPerfMon::Prim),
+				(int)pm.Get(GSPerfMon::Draw),
+				(int)std::ceil(pm.Get(GSPerfMon::DrawCalls)),
+				(int)std::ceil(pm.Get(GSPerfMon::DrawCallsROV)),
+				(int)std::ceil(pm.Get(GSPerfMon::Barriers)),
+				(int)std::ceil(pm.Get(GSPerfMon::BarriersROV)),
+				(int)std::ceil(pm.Get(GSPerfMon::RenderPasses)),
+				(int)std::ceil(pm.Get(GSPerfMon::Readbacks)),
+				(int)std::ceil(pm.Get(GSPerfMon::TextureCopies)),
+				(int)std::ceil(pm.Get(GSPerfMon::DepthCopiesROV)),
+				(int)std::ceil(pm.Get(GSPerfMon::TextureUploads)));
+		}
 	}
 }
 

pcsx2/GS/GSGL.h

@@ -46,10 +46,15 @@
 	#define GL_POP()      g_gs_device->PopDebugGroup()
 	#define GL_INS(...)   g_gs_device->InsertDebugMessage(GSDevice::DebugMessageCategory::Message, __VA_ARGS__)
 	#define GL_PERF(...)  g_gs_device->InsertDebugMessage(GSDevice::DebugMessageCategory::Performance, __VA_ARGS__)
+	#define GL_ROV(...) \
+		if (GSConfig.HWROVLogging) \
+			Console.Warning(__VA_ARGS__); \
+		g_gs_device->InsertDebugMessage(GSDevice::DebugMessageCategory::Message, __VA_ARGS__);
 #else
 	#define GL_PUSH_(...) (void)(0)
 	#define GL_PUSH(...) (void)(0)
 	#define GL_POP()     (void)(0)
 	#define GL_INS(...)  (void)(0)
 	#define GL_PERF(...) (void)(0)
+	#define GL_ROV(...) (void)(0)
 #endif

pcsx2/GS/GSPerfMon.h

@@ -23,6 +23,9 @@ public:
 		SyncPoint,
 		Barriers,
 		RenderPasses,
+		DepthCopiesROV, // Overlaps with regular texture copies.
+		DrawCallsROV, // Overlaps with regular draw calls.
+		BarriersROV, // Overlaps with regular barriers.
 		CounterLast,
 
 		// Reused counters for HW.

pcsx2/GS/GSRegs.h

@@ -203,8 +203,8 @@ enum GS_AFAIL
 
 enum GS_ALPHA_BITS
 {
-	ALPHA_ABC_CS = 0,
-	ALPHA_ABC_CD = 1,
+	ALPHA_ABD_CS = 0,
+	ALPHA_ABD_CD = 1,
 	ALPHA_C_AS   = 0,
 	ALPHA_C_AD   = 1,
 	ALPHA_C_FIX  = 2,

pcsx2/GS/GSState.cpp

@@ -4226,7 +4226,7 @@ bool GSState::TrianglesAreQuads(bool shuffle_check)
 }
 
 template<u32 primclass>
-GSState::PRIM_OVERLAP GSState::GetPrimitiveOverlapDrawlistImpl(bool save_drawlist, bool save_bbox, float bbox_scale)
+GSState::PRIM_OVERLAP GSState::GetPrimitiveOverlapDrawlistImpl(bool save_drawlist, bool save_bbox, float bbox_scale, u32* max_size)
 {
 	const GSVector4i xyof = m_context->scissor.xyof.xyxy();
 
@@ -4274,7 +4274,14 @@ GSState::PRIM_OVERLAP GSState::GetPrimitiveOverlapDrawlistImpl(bool save_drawlis
 	if (primclass == GS_TRIANGLE_CLASS && m_quad_check_valid && m_are_quads && !using_aa1)
 	{
 		// The triangles-are-quads check already ensures that there is no overlap.
-		m_drawlist.push_back(m_index->tail / n);
+		if (save_drawlist)
+		{
+			m_drawlist.push_back(m_index->tail / n);
+		}
+		else if (max_size)
+		{
+			*max_size = 1;
+		}
 		if (save_bbox)
 		{
 			const GSVector4i draw_area = GSVector4i(m_vt.m_min.p.upld(m_vt.m_max.p) * GSVector4(16.0f)) + xyof;
@@ -4286,6 +4293,7 @@ GSState::PRIM_OVERLAP GSState::GetPrimitiveOverlapDrawlistImpl(bool save_drawlis
 	PRIM_OVERLAP overlap = PRIM_OVERLAP_NO;
 	bool check_quads = (primclass == GS_TRIANGLE_CLASS) && !using_aa1;
 
+	u32 drawlist_size = 0;
 	u32 i = 0;
 	u32 skip = 0; // Number of indices to skip if we have the bbox from the previous iteration.
 	
@@ -4683,6 +4691,17 @@ GSState::PRIM_OVERLAP GSState::GetPrimitiveOverlapDrawlistImpl(bool save_drawlis
 		{
 			m_drawlist.push_back((j - i) / n); // Prim count
 		}
+		else if (max_size)
+		{
+			// If the max size pointer is passed it means we just want to peek at
+			// the drawlist size up to the given limit to avoid unecessary work.
+			drawlist_size++;
+			if (drawlist_size >= *max_size)
+			{
+				*max_size = drawlist_size;
+				return drawlist_size > 0 ? PRIM_OVERLAP_YES : PRIM_OVERLAP_UNKNOW;
+			}
+		}
 		else if (j < count)
 		{
 			return PRIM_OVERLAP_YES; // Early exit if not saving drawlist.
@@ -4696,21 +4715,27 @@ GSState::PRIM_OVERLAP GSState::GetPrimitiveOverlapDrawlistImpl(bool save_drawlis
 		all = bbox;
 		i = j;
 	}
+
+	if (max_size)
+	{
+		*max_size = drawlist_size;
+	}
+
 	return overlap;
 }
 
-GSState::PRIM_OVERLAP GSState::GetPrimitiveOverlapDrawlist(bool save_drawlist, bool save_bbox, float bbox_scale)
+GSState::PRIM_OVERLAP GSState::GetPrimitiveOverlapDrawlist(bool save_drawlist, bool save_bbox, float bbox_scale, u32* max_size)
 {
 	switch (m_vt.m_primclass)
 	{
 		case GS_POINT_CLASS:
-			return GetPrimitiveOverlapDrawlistImpl<GS_POINT_CLASS>(save_drawlist, save_bbox, bbox_scale);
+			return GetPrimitiveOverlapDrawlistImpl<GS_POINT_CLASS>(save_drawlist, save_bbox, bbox_scale, max_size);
 		case GS_LINE_CLASS:
-			return GetPrimitiveOverlapDrawlistImpl<GS_LINE_CLASS>(save_drawlist, save_bbox, bbox_scale);
+			return GetPrimitiveOverlapDrawlistImpl<GS_LINE_CLASS>(save_drawlist, save_bbox, bbox_scale, max_size);
 		case GS_TRIANGLE_CLASS:
-			return GetPrimitiveOverlapDrawlistImpl<GS_TRIANGLE_CLASS>(save_drawlist, save_bbox, bbox_scale);
+			return GetPrimitiveOverlapDrawlistImpl<GS_TRIANGLE_CLASS>(save_drawlist, save_bbox, bbox_scale, max_size);
 		case GS_SPRITE_CLASS:
-			return GetPrimitiveOverlapDrawlistImpl<GS_SPRITE_CLASS>(save_drawlist, save_bbox, bbox_scale);
+			return GetPrimitiveOverlapDrawlistImpl<GS_SPRITE_CLASS>(save_drawlist, save_bbox, bbox_scale, max_size);
 		default:
 			pxFail("Invalid primclass."); // Impossible.
 			return PRIM_OVERLAP_UNKNOW;

pcsx2/GS/GSState.h

@@ -533,8 +533,10 @@ public:
 	bool TrianglesAreQuadsImpl();
 	bool TrianglesAreQuads(bool shuffle_check = false);
 	template <u32 primclass>
-	PRIM_OVERLAP GetPrimitiveOverlapDrawlistImpl(bool save_drawlist = false, bool save_bbox = false, float bbox_scale = 1.0f);
-	PRIM_OVERLAP GetPrimitiveOverlapDrawlist(bool save_drawlist = false, bool save_bbox = false, float bbox_scale = 1.0f);
+	PRIM_OVERLAP GetPrimitiveOverlapDrawlistImpl(bool save_drawlist = false, bool save_bbox = false,
+		float bbox_scale = 1.0f, u32* max_size = nullptr);
+	PRIM_OVERLAP GetPrimitiveOverlapDrawlist(bool save_drawlist = false, bool save_bbox = false,
+		float bbox_scale = 1.0f, u32* max_size = nullptr);
 	PRIM_OVERLAP PrimitiveOverlap(bool save_drawlist = false);
 	template <u32 primclass, bool fst>
 	void GetPrimitiveOverlapDrawlistTextureBBoxImpl(float bbox_scale = 1.0f);

pcsx2/GS/Renderers/Common/GSDevice.cpp

@@ -693,6 +693,12 @@ void GSDevice::Recycle(GSTexture* t)
 
 	t->SetLastFrameUsed(m_frame);
 
+	t->ClearUnorderedAccess();
+	
+#ifdef PCSX2_DEVBUILD
+	t->SetDebugName("");
+#endif
+
 	FastList<GSTexture*>& pool = m_pool[!t->IsTexture()];
 	pool.push_front(t);
 	m_pool_memory_usage += t->GetMemUsage();
@@ -776,6 +782,18 @@ GSTexture* GSDevice::CreateDepthStencil(const GSVector2i& size, bool clear, bool
 		clear, !prefer_reuse);
 }
 
+GSTexture* GSDevice::CreateDepthColor(int w, int h, bool clear, bool prefer_reuse)
+{
+	return FetchSurface(GSTexture::Type::RenderTarget, w, h, 1, GSTexture::Format::DepthColor,
+		clear, !prefer_reuse);
+}
+
+GSTexture* GSDevice::CreateDepthColor(const GSVector2i& size, bool clear, bool prefer_reuse)
+{
+	return FetchSurface(GSTexture::Type::RenderTarget, size.x, size.y, 1, GSTexture::Format::DepthColor,
+		clear, !prefer_reuse);
+}
+
 GSTexture* GSDevice::CreateTexture(int w, int h, int mipmap_levels, GSTexture::Format format, bool prefer_reuse /* = false */)
 {
 	pxAssert(mipmap_levels != 0 && (mipmap_levels < 0 || mipmap_levels <= GetMipmapLevelsForSize(w, h)));
@@ -1486,9 +1504,9 @@ static const char* GetSetDATMName(SetDATM datm)
 	return "Unknown";
 }
 
-static const char* GetPSAA1Name(u32 aa1)
+static const char* GetPSAA1Name(GSHWDrawConfig::PS_AA1 aa1)
 {
-	switch (static_cast<GSHWDrawConfig::PS_AA1>(aa1))
+	switch (aa1)
 	{
 		case GSHWDrawConfig::PS_AA1::NONE: return "NONE";
 		case GSHWDrawConfig::PS_AA1::LINE: return "LINE";
@@ -1509,6 +1527,17 @@ static const char* GetTexHazardName(u32 tex_hazard)
 	return "Unknown";
 }
 
+static const char* GetPSROVDepthname(GSHWDrawConfig::PS_ROV_DEPTH rov_depth)
+{
+	switch (rov_depth)
+	{
+		case GSHWDrawConfig::PS_ROV_DEPTH::NONE: return "NONE";
+		case GSHWDrawConfig::PS_ROV_DEPTH::READ_ONLY: return "READ_ONLY";
+		case GSHWDrawConfig::PS_ROV_DEPTH::READ_WRITE: return "READ_WRITE";
+	}
+	return "Unknown";
+}
+
 static void DumpPSSelector(DrawConfigWriter& out, const GSHWDrawConfig::PSSelector& ps)
 {
 	out.WriteLn("aem_fmt: {}", ps.aem_fmt);
@@ -1564,9 +1593,13 @@ static void DumpPSSelector(DrawConfigWriter& out, const GSHWDrawConfig::PSSelect
 	out.WriteLn("point_sampler: {}", ps.point_sampler);
 	out.WriteLn("region_rect: {}", ps.region_rect);
 	out.WriteLn("scanmsk: {} ({})", GSUtil::GetSCANMSKName(ps.scanmsk), ps.scanmsk);
-	out.WriteLn("aa1: {} ({})", static_cast<u32>(ps.aa1), GetPSAA1Name(static_cast<u32>(ps.aa1)));
+	out.WriteLn("aa1: {} ({})", GetPSAA1Name(ps.aa1), static_cast<u32>(ps.aa1));
 	out.WriteLn("abe: {}", static_cast<u32>(ps.abe));
 	out.WriteLn("sw_aniso: {}", ps.sw_aniso);
+	out.WriteLn("rov_color: {}", ps.rov_color);
+	out.WriteLn("rov_depth: {} ({})", GetPSROVDepthname(ps.rov_depth), static_cast<u32>(ps.rov_depth));
+	out.WriteLn("ztst: {} ({})", GSUtil::GetZTSTName(ps.ztst), static_cast<u32>(ps.ztst));
+	out.WriteLn("zfloor: {}", static_cast<u32>(ps.zfloor));
 }
 
 static void DumpVSSelector(DrawConfigWriter& out, const GSHWDrawConfig::VSSelector& vs)
@@ -1690,6 +1723,8 @@ static void DumpConfig(DrawConfigWriter& out, const GSHWDrawConfig& conf,
 	out.WriteLn("topology: {} ({})", GetTopologyName(conf.topology), static_cast<u32>(conf.topology));
 	out.WriteLn("require_one_barrier: {}", conf.require_one_barrier);
 	out.WriteLn("require_full_barrier: {}", conf.require_full_barrier);
+	DumpVector4(out, "drawarea", conf.drawarea);
+	DumpVector4(out, "samplearea", conf.samplearea);
 	out.WriteLn("tex_hazard: {}", GetTexHazardName(conf.tex_hazard));
 
 	out.WriteLn("destination_alpha: {} ({})", GetDestinationAlphaModeName(conf.destination_alpha), static_cast<u32>(conf.destination_alpha));

pcsx2/GS/Renderers/Common/GSDevice.h

@@ -666,6 +666,7 @@ struct alignas(16) GSHWDrawConfig
 	using PS_ATST  = GSShader::PS_ATST;
 	using PS_AFAIL = GSShader::PS_AFAIL;
 	using PS_AA1   = GSShader::PS_AA1;
+	using PS_ROV_DEPTH = GSShader::PS_ROV_DEPTH;
 #pragma pack(push, 1)
 	struct VSSelector
 	{
@@ -786,12 +787,16 @@ struct alignas(16) GSHWDrawConfig
 
 				// Anisotropic filtering
 				u32 sw_aniso : 5;
+				
+				// ROVs
+				u32 rov_color : 1;
+				PS_ROV_DEPTH rov_depth : 2;
 			};
 
 			struct
 			{
 				u64 key_lo;
-				u32 key_hi;
+				u64 key_hi;
 			};
 		};
 		__fi PSSelector() : key_lo(0), key_hi(0) {}
@@ -800,6 +805,21 @@ struct alignas(16) GSHWDrawConfig
 		__fi bool operator!=(const PSSelector& rhs) const { return (key_lo != rhs.key_lo || key_hi != rhs.key_hi); }
 		__fi bool operator<(const PSSelector& rhs) const { return (key_lo < rhs.key_lo || key_hi < rhs.key_hi); }
 
+		__fi bool IsSWBlending() const
+		{
+			return blend_a || blend_b || blend_d;
+		}
+
+		__fi bool IsZTesting() const
+		{
+			return ztst == ZTST_GEQUAL || ztst == ZTST_GREATER;
+		}
+
+		__fi bool IsAlphaTesting() const
+		{
+			return atst != PS_ATST::NONE;
+		}
+
 		__fi bool IsFeedbackLoopRT() const
 		{
 			const u32 sw_blend_bits = blend_a | blend_b | blend_d;
@@ -816,6 +836,11 @@ struct alignas(16) GSHWDrawConfig
 			return afail_needs_depth || ztst_needs_depth || aa1_needs_depth;
 		}
 
+		__fi bool HasShaderDiscard() const
+		{
+			return (IsAlphaTesting() && afail == PS_AFAIL::KEEP) || scanmsk || date || IsZTesting();
+		}
+
 		/// Disables color output from the pixel shader, this is done when all channels are masked.
 		__fi void DisableColorOutput()
 		{
@@ -844,9 +869,39 @@ struct alignas(16) GSHWDrawConfig
 			{
 				aa1 = PS_AA1::TRIANGLE;
 			}
+
+			if (rov_depth == PS_ROV_DEPTH::READ_WRITE)
+			{
+				rov_depth = PS_ROV_DEPTH::READ_ONLY;
+			}
+		}
+
+		__fi bool HasColorOutput() const
+		{
+			return !no_color;
+		}
+
+		__fi bool HasDepthOutput() const
+		{
+			return zfloor || zclamp || IsFeedbackLoopDepth() || (rov_depth == PS_ROV_DEPTH::READ_WRITE);
+		}
+
+		__fi bool HasColorROV() const
+		{
+			return rov_color != 0;
+		}
+
+		__fi bool HasDepthROV() const
+		{
+			return rov_depth == PS_ROV_DEPTH::READ_ONLY || rov_depth == PS_ROV_DEPTH::READ_WRITE;
+		}
+
+		__fi bool HasDepthROVWrite() const
+		{
+			return rov_depth == PS_ROV_DEPTH::READ_WRITE;
 		}
 	};
-	static_assert(sizeof(PSSelector) == 12, "PSSelector is 12 bytes");
+	static_assert(sizeof(PSSelector) == 16, "PSSelector is 12 bytes");
 #pragma pack(pop)
 	struct PSSelectorHash
 	{
@@ -1255,6 +1310,11 @@ struct alignas(16) GSHWDrawConfig
 	{
 		return ps.IsFeedbackLoopDepth() || (tex_hazard == TEX_HAZARD_DEPTH);
 	}
+	
+	bool IsBlending()
+	{
+		return blend.enable || blend_multi_pass.enable || ps.IsSWBlending();
+	}
 
 	// Dumping
 	static void DumpConfig(const std::string& path, const GSHWDrawConfig& conf,
@@ -1330,6 +1390,7 @@ public:
 		bool test_and_sample_depth: 1; ///< Supports concurrently binding the depth-stencil buffer for sampling and depth testing.
 		bool depth_feedback       : 1; ///< Depth feedback loops can be done with DS directly (otherwise need to copy to separate RT).  Implies `feedback_loops`.
 		bool aa1                  : 1; ///< Supports the GS AA1 feature.
+		bool rov                  : 1; ///< Supports rasterizer ordered views for both depth and color.
 		FeatureSupport()
 		{
 			memset(this, 0, sizeof(*this));
@@ -1563,9 +1624,11 @@ public:
 
 	GSTexture* CreateRenderTarget(int w, int h, GSTexture::Format format, bool clear = true, bool prefer_reuse = true);
 	GSTexture* CreateDepthStencil(int w, int h, bool clear = true, bool prefer_reuse = true);
+	GSTexture* CreateDepthColor(int w, int h, bool clear = true, bool prefer_reuse = true);
 	GSTexture* CreateTexture(int w, int h, int mipmap_levels, GSTexture::Format format, bool prefer_reuse = false);
 	GSTexture* CreateRenderTarget(const GSVector2i& size, GSTexture::Format format, bool clear = true, bool prefer_reuse = true);
 	GSTexture* CreateDepthStencil(const GSVector2i& size, bool clear = true, bool prefer_reuse = true);
+	GSTexture* CreateDepthColor(const GSVector2i& size, bool clear = true, bool prefer_reuse = true);
 	GSTexture* CreateTexture(const GSVector2i& size, int mipmap_levels, GSTexture::Format format, bool prefer_reuse = false);
 	GSTexture* CreateCompatible(GSTexture* tex, bool clear = true, bool prefer_reuse = true);
 	GSTexture* CreateCompatible(GSTexture* tex, const GSVector2i& size, bool clear = true, bool prefer_reuse = true);

pcsx2/GS/Renderers/Common/GSShaderEnums.h

@@ -56,4 +56,11 @@ enum class PS_AA1 : uint32_t
 	TRIANGLE_SW_Z = 3, ///< AA1 triangles with software Z discard
 };
 
+enum class PS_ROV_DEPTH : uint32_t
+{
+	NONE = 0,
+	READ_WRITE = 1,
+	READ_ONLY = 2,
+};
+
 } // namespace GSShader

pcsx2/GS/Renderers/Common/GSTexture.h

@@ -73,6 +73,13 @@ protected:
 	bool m_needs_mipmaps_generated = true;
 	ClearValue m_clear_value = {};
 
+	// For GL/DX11 since they don't track layouts.
+	// Used heuristically to decide whether to use ROV for a draw.
+	bool m_unordered_access = false;
+
+#ifdef PCSX2_DEVBUILD
+	std::string m_debug_name;
+#endif
 public:
 	GSTexture();
 	virtual ~GSTexture();
@@ -87,6 +94,7 @@ public:
 
 #ifdef PCSX2_DEVBUILD
 	virtual void SetDebugName(std::string_view name) = 0;
+	const std::string& GetDebugName() { return m_debug_name; }
 #endif
 
 	bool Save(const std::string& fn);
@@ -173,6 +181,12 @@ public:
 	// Typical size of a RGBA texture
 	u32 GetMemUsage() const { return m_size.x * m_size.y * (m_format == Format::UNorm8 ? 1 : 4); }
 
+	// The unordered access flag is sticky, so once it's set we keep using ROV for the target
+	// until it's recycled/destroyed. Only used for DX11/GL, which don't track actual resource layout/state.
+	virtual bool IsUnorderedAccess() const { return m_unordered_access; }
+	void UseUnorderedAccess() { m_unordered_access = true; }
+	void ClearUnorderedAccess() { m_unordered_access = false; }
+
 	// Helper routines for formats/types
 	static bool IsCompressedFormat(Format format) { return (format >= Format::BC1 && format <= Format::BC7); }
 };

pcsx2/GS/Renderers/DX11/D3D11ShaderCache.cpp

@@ -65,6 +65,7 @@ bool D3D11ShaderCache::Open(D3D_FEATURE_LEVEL feature_level, bool debug)
 			m_shader_model = D3D::ShaderModel::SM41;
 			break;
 		case D3D_FEATURE_LEVEL_11_0:
+		case D3D_FEATURE_LEVEL_11_1:
 			m_shader_model = D3D::ShaderModel::SM50;
 			break;
 		default:

pcsx2/GS/Renderers/DX11/GSDevice11.cpp

@@ -109,7 +109,8 @@ bool GSDevice11::Create(GSVSyncMode vsync_mode, bool allow_present_throttle)
 
 	wil::com_ptr_nothrow<IDXGIAdapter1> dxgi_adapter = D3D::GetAdapterByName(m_dxgi_factory.get(), GSConfig.Adapter);
 
-	static constexpr std::array<D3D_FEATURE_LEVEL, 2> requested_feature_levels = {{
+	static constexpr std::array<D3D_FEATURE_LEVEL, 3> requested_feature_levels = {{
+		D3D_FEATURE_LEVEL_11_1,
 		D3D_FEATURE_LEVEL_11_0,
 		D3D_FEATURE_LEVEL_10_0,
 	}};
@@ -592,11 +593,18 @@ bool GSDevice11::Create(GSVSyncMode vsync_mode, bool allow_present_throttle)
 			Host::OSD_WARNING_DURATION);
 	}
 
+	// 1x1 dummy texture.
+	m_null_texture = CreateSurface(GSTexture::Type::RenderTarget, 1, 1, 1, GSTexture::Format::Color);
+	if (!m_null_texture)
+		return false;
+
 	return true;
 }
 
 void GSDevice11::Destroy()
 {
+	delete m_null_texture;
+	
 	GSDevice::Destroy();
 	DestroySwapChain();
 	DestroyTimestampQueries();
@@ -683,6 +691,12 @@ void GSDevice11::SetFeatures(IDXGIAdapter1* adapter)
 	m_conservative_depth = (m_feature_level >= D3D_FEATURE_LEVEL_11_0);
 	m_rgba16_unorm_hw_blend = IsTextureFormatHWBlendable(m_dev.get(), DXGI_FORMAT_R16G16B16A16_UNORM);
 
+	m_uav_texture = (m_feature_level >= D3D_FEATURE_LEVEL_11_0);
+
+	D3D11_FEATURE_DATA_D3D11_OPTIONS2 options2{};
+	m_dev->CheckFeatureSupport(D3D11_FEATURE_D3D11_OPTIONS2, &options2, sizeof(options2));
+	m_features.rov = m_uav_texture && options2.ROVsSupported;
+
 	// Let the user know if said features are available.
 	Console.WriteLnFmt("D3D11: DXTn Texture Compression: {}", m_features.dxt_textures ? "Supported" : "Not Supported");
 	Console.WriteLnFmt("D3D11: BC6/7 Texture Compression: {}", m_features.bptc_textures ? "Supported" : "Not Supported");
@@ -1322,6 +1336,8 @@ GSTexture* GSDevice11::CreateSurface(GSTexture::Type type, int width, int height
 	{
 		case GSTexture::Type::RenderTarget:
 			desc.BindFlags = D3D11_BIND_RENDER_TARGET | D3D11_BIND_SHADER_RESOURCE;
+			if (m_uav_texture)
+				desc.BindFlags |= D3D11_BIND_UNORDERED_ACCESS;
 			break;
 		case GSTexture::Type::DepthStencil:
 			desc.BindFlags = D3D11_BIND_DEPTH_STENCIL | D3D11_BIND_SHADER_RESOURCE;
@@ -1331,6 +1347,7 @@ GSTexture* GSDevice11::CreateSurface(GSTexture::Type type, int width, int height
 			desc.MiscFlags = (levels > 1 && !GSTexture::IsCompressedFormat(format)) ? D3D11_RESOURCE_MISC_GENERATE_MIPS : 0;
 			break;
 		case GSTexture::Type::RWTexture:
+			pxAssert(m_uav_texture);
 			desc.BindFlags = D3D11_BIND_UNORDERED_ACCESS | D3D11_BIND_SHADER_RESOURCE;
 			break;
 		default:
@@ -1482,7 +1499,7 @@ void GSDevice11::DoStretchRect(GSTexture* sTex, const GSVector4& sRect, GSTextur
 
 	// ps
 
-	PSSetShaderResource(0, sTex);
+	PSSetShaderResource(TEXTURE_TEXTURE, sTex);
 	PSSetSamplerState(filter == Biln ? m_convert.ln.get() : m_convert.pt.get());
 	PSSetShader(ps, ps_cb);
 
@@ -1547,7 +1564,7 @@ void GSDevice11::PresentRect(GSTexture* sTex, const GSVector4& sRect, GSTexture*
 
 	// ps
 
-	PSSetShaderResource(0, sTex);
+	PSSetShaderResource(TEXTURE_TEXTURE, sTex);
 	PSSetSamplerState(filter == Biln ? m_convert.ln.get() : m_convert.pt.get());
 	PSSetShader(m_present.ps[static_cast<u32>(shader)].get(), m_present.ps_cb.get());
 
@@ -1698,7 +1715,7 @@ void GSDevice11::DoMultiStretchRects(const MultiStretchRect* rects, u32 num_rect
 	IASetIndexBuffer(m_ib.get());
 
 	CommitClear(rects[0].src);
-	PSSetShaderResource(0, rects[0].src);
+	PSSetShaderResource(TEXTURE_TEXTURE, rects[0].src);
 	PSSetSamplerState(rects[0].filter == Biln ? m_convert.ln.get() : m_convert.pt.get());
 
 	OMSetBlendState(m_convert.bs[rects[0].wmask.wrgba].get(), 0.0f);
@@ -1921,6 +1938,8 @@ void GSDevice11::SetupPS(const PSSelector& sel, const GSHWDrawConfig::PSConstant
 		sm.AddMacro("PS_AA1", static_cast<u32>(sel.aa1));
 		sm.AddMacro("PS_ABE", sel.abe);
 		sm.AddMacro("PS_ANISOTROPIC_FILTERING", sel.sw_aniso);
+		sm.AddMacro("PS_ROV_COLOR", sel.rov_color);
+		sm.AddMacro("PS_ROV_DEPTH", static_cast<u32>(sel.rov_depth));
 
 		wil::com_ptr_nothrow<ID3D11PixelShader> ps = m_shader_cache.GetPixelShader(m_dev.get(), m_tfx_source, sm.GetPtr(), "ps_main");
 		i = m_ps.try_emplace(sel, std::move(ps)).first;
@@ -2339,7 +2358,7 @@ void GSDevice11::SetupDATE(GSTexture* rt, GSTexture* ds, SetDATM datm, const GSV
 
 	// ps
 
-	PSSetShaderResource(0, rt);
+	PSSetShaderResource(TEXTURE_TEXTURE, rt);
 	PSSetSamplerState(m_convert.pt.get());
 	PSSetShader(GetConvertShader(SetDATMShader(datm)), nullptr);
 
@@ -2707,10 +2726,16 @@ void GSDevice11::OMSetBlendState(ID3D11BlendState* bs, u8 bf)
 	}
 }
 
-void GSDevice11::OMSetRenderTargets(GSTexture* rt, GSTexture* ds, const GSVector4i* scissor, ID3D11DepthStencilView* read_only_dsv)
+void GSDevice11::OMSetRenderTargets(GSTexture* rt, GSTexture* ds, GSTexture* rt_uav_tex, GSTexture* ds_uav_tex,
+	const GSVector4i* scissor, ID3D11DepthStencilView* read_only_dsv)
 {
+	if (!(rt || rt_uav_tex) && ds_uav_tex)
+		rt_uav_tex = m_null_texture; // Fill in the first UAV slot with the null texture.
+
 	ID3D11RenderTargetView* rtv = nullptr;
 	ID3D11DepthStencilView* dsv = nullptr;
+	ID3D11UnorderedAccessView* rt_uav = nullptr;
+	ID3D11UnorderedAccessView* ds_uav = nullptr;
 
 	if (rt)
 	{
@@ -2722,9 +2747,21 @@ void GSDevice11::OMSetRenderTargets(GSTexture* rt, GSTexture* ds, const GSVector
 		CommitClear(ds);
 		dsv = read_only_dsv ? read_only_dsv : *static_cast<GSTexture11*>(ds);
 	}
+	if (rt_uav_tex)
+	{
+		CommitClear(rt_uav_tex);
+		rt_uav = *static_cast<GSTexture11*>(rt_uav_tex);
+	}
+	if (ds_uav_tex)
+	{
+		CommitClear(ds_uav_tex);
+		ds_uav = *static_cast<GSTexture11*>(ds_uav_tex);
+	}
 
-	const bool changed = (m_state.rtv != rtv || m_state.dsv != dsv);
-	g_perfmon.Put(GSPerfMon::RenderPasses, static_cast<double>(changed));
+	const bool changed = (m_state.rtv != rtv || m_state.dsv != dsv || m_state.rt_uav != rt_uav || m_state.ds_uav != ds_uav);
+
+	if (changed)
+		g_perfmon.Put(GSPerfMon::RenderPasses, 1.0);
 
 	if (m_state.rtv != rtv)
 	{
@@ -2744,15 +2781,49 @@ void GSDevice11::OMSetRenderTargets(GSTexture* rt, GSTexture* ds, const GSVector
 		m_state.dsv = dsv;
 		m_state.current_ds = ds;
 	}
+	if (m_state.rt_uav != rt_uav)
+	{
+		if (m_state.rt_uav)
+			m_state.rt_uav->Release();
+		if (rt_uav)
+			rt_uav->AddRef();
+		m_state.rt_uav = rt_uav;
+		m_state.current_rt_uav = rt_uav_tex;
+	}
+	if (m_state.ds_uav != ds_uav)
+	{
+		if (m_state.ds_uav)
+			m_state.ds_uav->Release();
+		if (ds_uav)
+			ds_uav->AddRef();
+		m_state.ds_uav = ds_uav;
+		m_state.current_ds_uav = ds_uav_tex;
+	}
 
 	PSUnbindConflictingSRVs(m_state.current_rt, read_only_dsv ? nullptr : m_state.current_ds);
+	PSUnbindConflictingSRVs(m_state.current_rt_uav, m_state.current_ds_uav);
 
 	if (changed)
-		m_ctx->OMSetRenderTargets(1, &rtv, dsv);
-
-	if (rt || ds)
 	{
-		const GSVector2i size = rt ? rt->GetSize() : ds->GetSize();
+		// OM targets and UAVs share the same namespace in DX11, so we need to pack them into contiguous slots.
+		u32 num_rtvs = rt ? 1 : 0;
+		u32 num_uavs = 0;
+		ID3D11UnorderedAccessView* uavs[2];
+		if (rt_uav)
+			uavs[num_uavs++] = rt_uav;
+		if (ds_uav)
+			uavs[num_uavs++] = ds_uav;
+		ID3D11RenderTargetView* rtvs[] = { rtv };
+		m_ctx->OMSetRenderTargetsAndUnorderedAccessViews(num_rtvs, rtvs, dsv, num_rtvs, num_uavs, uavs, nullptr);
+	}
+
+	if (rt || ds || rt_uav_tex || ds_uav_tex)
+	{
+		const GSVector2i size =
+			rt ? rt->GetSize() :
+			ds ? ds->GetSize() :
+			(rt_uav_tex && rt_uav_tex != m_null_texture) ? rt_uav_tex->GetSize() :
+			ds_uav_tex->GetSize();
 		SetViewport(size);
 		SetScissor(scissor ? *scissor : GSVector4i::loadh(size));
 	}
@@ -2815,6 +2886,10 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 {
 	const GSVector2i rtsize = (config.rt ? config.rt : config.ds)->GetSize();
 	GSTexture* colclip_rt = g_gs_device->GetColorClipTexture();
+	GSTexture* draw_rt = config.ps.HasColorROV() ? nullptr : config.rt;
+	GSTexture* draw_ds = config.ps.HasDepthROV() ? nullptr : config.ds;
+	GSTexture* draw_rt_rov = config.ps.HasColorROV() ? config.rt : nullptr;
+	GSTexture* draw_ds_rov = config.ps.HasDepthROV() ? config.ds : nullptr;
 	GSTexture* draw_rt_clone = nullptr;
 	GSTexture* draw_ds_clone = nullptr;
 	GSTexture* primid_texture = nullptr;
@@ -2865,6 +2940,8 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 			const GSVector4 sRect = dRect / GSVector4(rtsize.x, rtsize.y).xyxy();
 			StretchRect(config.rt, sRect, colclip_rt, dRect, ShaderConvert::COLCLIP_INIT, Nearest);
 		}
+
+		draw_rt = colclip_rt ? colclip_rt : draw_rt;
 	}
 
 	// Destination Alpha Setup
@@ -2950,7 +3027,7 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 		const OMBlendSelector blend(GSHWDrawConfig::ColorMaskSelector(1),
 			GSHWDrawConfig::BlendState(true, CONST_ONE, CONST_ONE, 3 /* MIN */, CONST_ONE, CONST_ZERO, false, 0));
 		SetupOM(dss, blend, 0);
-		OMSetRenderTargets(primid_texture, config.ds, &config.scissor, read_only_dsv);
+		OMSetRenderTargets(primid_texture, config.ds, nullptr, nullptr, &config.scissor, read_only_dsv);
 		SetRenderHWShaderResources(config, nullptr);
 		Draw(config);
 
@@ -2960,16 +3037,14 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 	}
 
 	// Avoid changing framebuffer just to switch from rt+depth to rt and vice versa.
-	GSTexture* draw_rt = colclip_rt ? colclip_rt : config.rt;
-	GSTexture* draw_ds = config.ds;
 	// Make sure no tex is bound as both rtv and srv at the same time.
 	// All conflicts should've been taken care of by PSUnbindConflictingSRVs.
-	if (!draw_rt && draw_ds && m_state.rtv && m_state.current_rt && m_state.rtv == *static_cast<GSTexture11*>(m_state.current_rt) &&
+	if (!(draw_rt || draw_rt_rov) && draw_ds && m_state.rtv && m_state.current_rt && m_state.rtv == *static_cast<GSTexture11*>(m_state.current_rt) &&
 		m_state.current_ds == draw_ds && config.tex != m_state.current_rt && m_state.current_rt->GetSize() == draw_ds->GetSize())
 	{
 		draw_rt = m_state.current_rt;
 	}
-	else if (!draw_ds && draw_rt && m_state.dsv && m_state.current_ds && m_state.dsv == *static_cast<GSTexture11*>(m_state.current_ds) &&
+	else if (!(draw_ds || draw_ds_rov) && draw_rt && m_state.dsv && m_state.current_ds && m_state.dsv == *static_cast<GSTexture11*>(m_state.current_ds) &&
 		m_state.current_rt == draw_rt && config.tex != m_state.current_ds && m_state.current_ds->GetSize() == draw_rt->GetSize())
 	{
 		draw_ds = m_state.current_ds;
@@ -2977,7 +3052,7 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 
 	const bool rt_feedbackloop_pass1 = config.IsFeedbackLoopRT(config.ps);
 	const bool rt_feedbackloop_pass2 = config.IsFeedbackLoopRT(config.alpha_second_pass.ps);
-	if (draw_rt && (((config.require_one_barrier || (config.require_full_barrier && m_features.multidraw_fb_copy)) &&
+	if (draw_rt && !draw_rt_rov && (((config.require_one_barrier || (config.require_full_barrier && m_features.multidraw_fb_copy)) &&
 		(rt_feedbackloop_pass1 || rt_feedbackloop_pass2))))
 	{
 		// Requires a copy of the RT.
@@ -2989,7 +3064,7 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 
 	const bool ds_feedbackloop_pass1 = config.IsFeedbackLoopDepth(config.ps);
 	const bool ds_feedbackloop_pass2 = config.IsFeedbackLoopDepth(config.alpha_second_pass.ps);
-	if (draw_ds && m_features.depth_feedback && (config.require_one_barrier || (config.require_full_barrier && m_features.multidraw_fb_copy)) &&
+	if (draw_ds && !draw_ds_rov && m_features.depth_feedback && (config.require_one_barrier || (config.require_full_barrier && m_features.multidraw_fb_copy)) &&
 		(ds_feedbackloop_pass1 || ds_feedbackloop_pass2))
 	{
 		// Requires a copy of the DS.
@@ -2999,7 +3074,7 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 			Console.Warning("D3D11: Failed to allocate temp texture for DS copy.");
 	}
 
-	OMSetRenderTargets(draw_rt, draw_ds, &config.scissor, read_only_dsv);
+	OMSetRenderTargets(draw_rt, draw_ds, draw_rt_rov, draw_ds_rov, &config.scissor, read_only_dsv);
 	SetRenderHWShaderResources(config, primid_texture);
 	SetupOM(config.depth, OMBlendSelector(config.colormask, config.blend), config.blend.constant);
 
@@ -3008,7 +3083,7 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 		m_ctx->ClearDepthStencilView(*static_cast<GSTexture11*>(draw_ds), D3D11_CLEAR_STENCIL, 0.0f, 1);
 
 	SendHWDraw(config, rt_feedbackloop_pass1 ? draw_rt_clone : nullptr, draw_rt, ds_feedbackloop_pass1 ? draw_ds_clone : nullptr, draw_ds,
-		config.require_one_barrier, config.require_full_barrier);
+		config.require_one_barrier, config.require_full_barrier, rtsize);
 
 	if (config.blend_multi_pass.enable)
 	{
@@ -3036,7 +3111,7 @@ void GSDevice11::RenderHW(GSHWDrawConfig& config)
 		const bool one_barrier = config.alpha_second_pass.require_one_barrier && m_features.multidraw_fb_copy;
 		SetupOM(config.alpha_second_pass.depth, OMBlendSelector(config.alpha_second_pass.colormask, config.blend), config.blend.constant);
 		SendHWDraw(config, rt_feedbackloop_pass2 ? draw_rt_clone : nullptr, draw_rt, ds_feedbackloop_pass2 ? draw_ds_clone : nullptr, draw_ds,
-			one_barrier, config.alpha_second_pass.require_full_barrier);
+			one_barrier, config.alpha_second_pass.require_full_barrier, rtsize);
 	}
 
 	if (colclip_rt)
@@ -3110,7 +3185,7 @@ void GSDevice11::FeedbackCopyAndBind(const GSHWDrawConfig& config,
 
 void GSDevice11::SendHWDraw(const GSHWDrawConfig& config,
 	GSTexture* draw_rt_clone, GSTexture* draw_rt, GSTexture* draw_ds_clone, GSTexture* draw_ds,
-	const bool one_barrier, const bool full_barrier)
+	const bool one_barrier, const bool full_barrier, GSVector2i rtsize)
 {
 #ifdef PCSX2_DEVBUILD
 	if ((one_barrier || full_barrier) && !(config.IsFeedbackLoopRT(config.ps) || config.IsFeedbackLoopDepth(config.ps))) [[unlikely]]
@@ -3149,6 +3224,9 @@ void GSDevice11::SendHWDraw(const GSHWDrawConfig& config,
 	}
 
 	Draw(config);
+
+	if (config.ps.HasColorROV() || config.ps.HasDepthROV())
+		g_perfmon.Put(GSPerfMon::DrawCallsROV, 1);
 }
 
 void GSDevice11::SetRenderHWShaderResources(const GSHWDrawConfig& config, GSTexture* primid_texture)
@@ -3156,13 +3234,13 @@ void GSDevice11::SetRenderHWShaderResources(const GSHWDrawConfig& config, GSText
 	if (config.tex && config.tex_hazard == GSHWDrawConfig::TEX_HAZARD_NONE)
 	{
 		CommitClear(config.tex);
-		PSSetShaderResource(0, config.tex);
+		PSSetShaderResource(TEXTURE_TEXTURE, config.tex);
 	}
 	if (config.pal)
 	{
 		CommitClear(config.pal);
-		PSSetShaderResource(1, config.pal);
+		PSSetShaderResource(TEXTURE_PALETTE, config.pal);
 	}
 	if (primid_texture)
-		PSSetShaderResource(3, primid_texture);
+		PSSetShaderResource(TEXTURE_PRIMID, primid_texture);
 }