From 40d5c785735670c4c6c7f694a7a0620158822bf2 Mon Sep 17 00:00:00 2001
From: lightningterror <18107717+lightningterror@users.noreply.github.com>
Date: Tue, 20 Aug 2024 15:40:50 +0200
Subject: [PATCH] GS/HW: Implement PABE(Per pixel alpha blending) on
 accumulation blend and add optimizations.

PABE accumulation blend:
Idea is to achieve final output Cs when As < 1, we do this with manipulating Cd using the src1 output.
This can't be done with reverse subtraction as we want Cd to be 0 when As < 1.
Blend mix is excluded as no games were found, otherwise it can be added.

PABE Disable blending:
We can disable blending here as an optimization since alpha max is 128
which if alpha is 1 in the formula Cs*Alpha + Cd*(1 - Alpha) will give us a result of Cs.
---
 bin/resources/shaders/dx11/tfx.fx        |  6 ++++
 bin/resources/shaders/opengl/tfx_fs.glsl |  6 ++++
 bin/resources/shaders/vulkan/tfx.glsl    |  6 ++++
 pcsx2/GS/Renderers/HW/GSRendererHW.cpp   | 41 ++++++++++++++++++++----
 pcsx2/GS/Renderers/Metal/tfx.metal       |  6 ++++
 pcsx2/ShaderCacheVersion.h               |  2 +-
 6 files changed, 60 insertions(+), 7 deletions(-)

diff --git a/bin/resources/shaders/dx11/tfx.fx b/bin/resources/shaders/dx11/tfx.fx
index 47b7a0478e..fb6811677b 100644
--- a/bin/resources/shaders/dx11/tfx.fx
+++ b/bin/resources/shaders/dx11/tfx.fx
@@ -862,9 +862,15 @@ void ps_blend(inout float4 Color, inout float4 As_rgba, float2 pos_xy)
 		// PABE
 		if (PS_PABE)
 		{
+			// As_rgba needed for accumulation blend to manipulate Cd.
 			// No blending so early exit
 			if (As < 1.0f)
+			{
+				As_rgba.rgb = (float3)0.0f;
 				return;
+			}
+
+			As_rgba.rgb = (float3)1.0f;
 		}
 
 		float4 RT = SW_BLEND_NEEDS_RT ? RtTexture.Load(int3(pos_xy, 0)) : (float4)0.0f;
diff --git a/bin/resources/shaders/opengl/tfx_fs.glsl b/bin/resources/shaders/opengl/tfx_fs.glsl
index fa53825d1e..dd3f8cb745 100644
--- a/bin/resources/shaders/opengl/tfx_fs.glsl
+++ b/bin/resources/shaders/opengl/tfx_fs.glsl
@@ -783,9 +783,15 @@ float As = As_rgba.a;
 
 	// PABE
 #if PS_PABE
+	// As_rgba needed for accumulation blend to manipulate Cd.
 	// No blending so early exit
 	if (As < 1.0f)
+	{
+		As_rgba.rgb = vec3(0.0f);
 		return;
+	}
+
+	As_rgba.rgb = vec3(1.0f);
 #endif
 
 #if SW_BLEND_NEEDS_RT
diff --git a/bin/resources/shaders/vulkan/tfx.glsl b/bin/resources/shaders/vulkan/tfx.glsl
index 2cc8b1e07d..70661deb13 100644
--- a/bin/resources/shaders/vulkan/tfx.glsl
+++ b/bin/resources/shaders/vulkan/tfx.glsl
@@ -1050,9 +1050,15 @@ void ps_blend(inout vec4 Color, inout vec4 As_rgba)
 
 		// PABE
 		#if PS_PABE
+			// As_rgba needed for accumulation blend to manipulate Cd
 			// No blending so early exit
 			if (As < 1.0f)
+			{
+				As_rgba.rgb = vec3(0.0f);
 				return;
+			}
+
+			As_rgba.rgb = vec3(1.0f);
 		#endif
 
 		#if PS_FEEDBACK_LOOP_IS_NEEDED
diff --git a/pcsx2/GS/Renderers/HW/GSRendererHW.cpp b/pcsx2/GS/Renderers/HW/GSRendererHW.cpp
index daa7d4ed15..33f147ae9d 100644
--- a/pcsx2/GS/Renderers/HW/GSRendererHW.cpp
+++ b/pcsx2/GS/Renderers/HW/GSRendererHW.cpp
@@ -4420,13 +4420,35 @@ void GSRendererHW::EmulateBlending(int rt_alpha_min, int rt_alpha_max, const boo
 	// Per pixel alpha blending
 	if (m_draw_env->PABE.PABE && GetAlphaMinMax().min < 128)
 	{
-		// Breath of Fire Dragon Quarter, Strawberry Shortcake, Super Robot Wars, Cartoon Network Racing, Simple 2000 Series Vol.81, SOTC, Super Robot Wars.
-		if (sw_blending)
+		// Breath of Fire Dragon Quarter, Strawberry Shortcake, Super Robot Wars, Cartoon Network Racing, Simple 2000 Series Vol.81, SOTC.
+		if (GetAlphaMinMax().max == 128 && m_conf.ps.blend_a == 0 && ((blend.dst == GSDevice::INV_SRC1_COLOR)
+			|| (blend.dst == GSDevice::INV_DST_ALPHA)
+			|| (blend.dst == GSDevice::INV_CONST_COLOR)))
 		{
-			GL_INS("PABE mode ENABLED");
-			if (features.texture_barrier)
+			// PABE disable blending:
+			// We can disable blending here as an optimization since alpha max is 128
+			// which if alpha is 1 in the formula Cs*Alpha + Cd*(1 - Alpha) will give us a result of Cs.
+			m_conf.blend = {};
+			m_conf.ps.no_color1 = true;
+			m_conf.ps.blend_a = m_conf.ps.blend_b = m_conf.ps.blend_c = m_conf.ps.blend_d = 0;
+
+			return;
+		}
+		else if (sw_blending)
+		{
+			if (accumulation_blend && (blend.op != GSDevice::OP_REV_SUBTRACT))
 			{
-				// Disable hw/sw blend and do pure sw blend with reading the framebuffer.
+				// PABE accumulation blend:
+				// Idea is to achieve final output Cs when As < 1, we do this with manipulating Cd using the src1 output.
+				// This can't be done with reverse subtraction as we want Cd to be 0 when As < 1.
+				// TODO: Blend mix is excluded as no games were found, otherwise it can be added.
+
+				m_conf.ps.pabe = 1;
+			}
+			else if (features.texture_barrier)
+			{
+				// PABE sw blend:
+				// Disable hw/sw mix and do pure sw blend with reading the framebuffer.
 				color_dest_blend   = false;
 				accumulation_blend = false;
 				blend_mix          = false;
@@ -4441,8 +4463,11 @@ void GSRendererHW::EmulateBlending(int rt_alpha_min, int rt_alpha_max, const boo
 			}
 			else
 			{
+				// PABE sw blend:
 				m_conf.ps.pabe = !(accumulation_blend || blend_mix);
 			}
+
+			GL_INS("PABE mode %s", m_conf.ps.pabe ? "ENABLED" : "DISABLED");
 		}
 	}
 
@@ -4516,9 +4541,13 @@ void GSRendererHW::EmulateBlending(int rt_alpha_min, int rt_alpha_max, const boo
 					m_conf.ps.blend_b = 2;
 				}
 			}
+			else if (m_conf.ps.pabe)
+			{
+				m_conf.blend.dst_factor = GSDevice::SRC1_COLOR;
+			}
 
 			// Dual source output not needed (accumulation blend replaces it with ONE).
-			m_conf.ps.no_color1 = true;
+			m_conf.ps.no_color1 = (m_conf.ps.pabe == 0);
 		}
 		else if (blend_mix)
 		{
diff --git a/pcsx2/GS/Renderers/Metal/tfx.metal b/pcsx2/GS/Renderers/Metal/tfx.metal
index fabe2653aa..a13c6cdb30 100644
--- a/pcsx2/GS/Renderers/Metal/tfx.metal
+++ b/pcsx2/GS/Renderers/Metal/tfx.metal
@@ -925,9 +925,15 @@ struct PSMain
 			// PABE
 			if (PS_PABE)
 			{
+				// As_rgba needed for accumulation blend to manipulate Cd.
 				// No blending so early exit
 				if (As < 1.f)
+				{
+					As_rgba.rgb = float3(0.f);
 					return;
+				}
+
+				As_rgba.rgb = float3(1.f);
 			}
 
 			float Ad = PS_RTA_CORRECTION ? trunc(current_color.a * 128.1f) / 128.f : trunc(current_color.a * 255.1f) / 128.f;
diff --git a/pcsx2/ShaderCacheVersion.h b/pcsx2/ShaderCacheVersion.h
index 44d799958f..f157d82388 100644
--- a/pcsx2/ShaderCacheVersion.h
+++ b/pcsx2/ShaderCacheVersion.h
@@ -3,4 +3,4 @@
 
 /// Version number for GS and other shaders. Increment whenever any of the contents of the
 /// shaders change, to invalidate the cache.
-static constexpr u32 SHADER_CACHE_VERSION = 55;
+static constexpr u32 SHADER_CACHE_VERSION = 56;