From ebe64c60e9c557645a5e6f44b9d3de3c2d2483f6 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Thu, 8 Aug 2019 11:49:35 +0200
Subject: [PATCH] Vulkan: Improve the Metal Gear Acid 2 performance hack.

---
 assets/compat.ini                       | 26 +++++----
 ext/native/thin3d/VulkanQueueRunner.cpp | 74 +++++++++++++++++++++++++
 2 files changed, 88 insertions(+), 12 deletions(-)

diff --git a/assets/compat.ini b/assets/compat.ini
index f1917c588254..74e11597fde1 100644
--- a/assets/compat.ini
+++ b/assets/compat.ini
@@ -403,13 +403,13 @@ NPJH50036 = true
 
 # Burnout games have problems with this on Mali, and have no use for it
 # Legends
-ULES00125 = true
-ULUS10025 = true
-ULJM05228 = true
-NPJH50305 = true
-ULJM05049 = true
-ULKS46027 = true
-ULAS42019 = true
+#ULES00125 = true
+#ULUS10025 = true
+#ULJM05228 = true
+#NPJH50305 = true
+#ULJM05049 = true
+#ULKS46027 = true
+#ULAS42019 = true
 
 # Dominator
 ULUS10236 = true
@@ -441,15 +441,17 @@ ULES01376 = true
 ULUS10466 = true
 
 [MGS2AcidHack]
+ULUS10006 = true # Metal Gear Acid
 ULES00008 = true
+ULJM05001 = true
+ULAS42007 = true
+ULJM08001 = true
+ULUS10077 = true # Metal Gear Acid 2
+ULAS42035 = true
 ULES00284 = true
 ULJM05047 = true
 ULKS46065 = true
-ULJM08001 = true
-ULJM05001 = true
-ULAS42007 = true
-ULUS10006 = true
-ULUS10077 = true
+ULJM08011 = true
 
 [SonicRivalsHack]
 ULES00622 = true  # SR1
diff --git a/ext/native/thin3d/VulkanQueueRunner.cpp b/ext/native/thin3d/VulkanQueueRunner.cpp
index a0042faf4e64..1a36d139cf18 100644
--- a/ext/native/thin3d/VulkanQueueRunner.cpp
+++ b/ext/native/thin3d/VulkanQueueRunner.cpp
@@ -463,6 +463,8 @@ void VulkanQueueRunner::RunSteps(VkCommandBuffer cmd, std::vector<VKRStep *> &st
 }
 
 void VulkanQueueRunner::ApplyMGSHack(std::vector<VKRStep *> &steps) {
+	// Really need a sane way to express transforms of steps.
+
 	// We want to turn a sequence of copy,render(1),copy,render(1),copy,render(1) to copy,copy,copy,render(n).
 
 	for (int i = 0; i < (int)steps.size() - 3; i++) {
@@ -530,6 +532,78 @@ void VulkanQueueRunner::ApplyMGSHack(std::vector<VKRStep *> &steps) {
 			break;
 		}
 	}
+
+	// There's also a post processing effect using depals that's just brutal in some parts
+	// of the game.
+	for (int i = 0; i < (int)steps.size() - 3; i++) {
+		int last = -1;
+		if (!(steps[i]->stepType == VKRStepType::RENDER &&
+			steps[i + 1]->stepType == VKRStepType::RENDER &&
+			steps[i + 2]->stepType == VKRStepType::RENDER &&
+			steps[i]->render.numDraws == 1 &&
+			steps[i + 1]->render.numDraws == 1 &&
+			steps[i + 2]->render.numDraws == 1 &&
+			steps[i]->render.color == VKRRenderPassAction::DONT_CARE &&
+			steps[i + 1]->render.color == VKRRenderPassAction::KEEP &&
+			steps[i + 2]->render.color == VKRRenderPassAction::DONT_CARE))
+			continue;
+		VKRFramebuffer *depalFramebuffer = steps[i]->render.framebuffer;
+		VKRFramebuffer *targetFramebuffer = steps[i + 1]->render.framebuffer;
+		// OK, found the start of a post-process sequence. Let's scan until we find the end.
+		for (int j = i; j < steps.size() - 3; j++) {
+			if (((j - i) & 1) == 0) {
+				// This should be a depal draw.
+				if (steps[j]->render.numDraws != 1)
+					break;
+				if (steps[j]->render.color != VKRRenderPassAction::DONT_CARE)
+					break;
+				if (steps[j]->render.framebuffer != depalFramebuffer)
+					break;
+				last = j;
+			} else {
+				// This should be a target draw.
+				if (steps[j]->render.numDraws != 1)
+					break;
+				if (steps[j]->render.color != VKRRenderPassAction::KEEP)
+					break;
+				if (steps[j]->render.framebuffer != targetFramebuffer)
+					break;
+				last = j;
+			}
+		}
+
+		if (last == -1)
+			continue;
+
+		// Combine the depal renders.
+		for (int j = i + 2; j <= last + 1; j += 2) {
+			for (int k = 0; k < (int)steps[j]->commands.size(); k++) {
+				switch (steps[j]->commands[k].cmd) {
+				case VKRRenderCommand::DRAW:
+				case VKRRenderCommand::DRAW_INDEXED:
+					steps[i]->commands.push_back(steps[j]->commands[k]);
+					break;
+				}
+			}
+			steps[j]->stepType = VKRStepType::RENDER_SKIP;
+		}
+
+		// Combine the target renders.
+		for (int j = i + 3; j <= last; j += 2) {
+			for (int k = 0; k < (int)steps[j]->commands.size(); k++) {
+				switch (steps[j]->commands[k].cmd) {
+				case VKRRenderCommand::DRAW:
+				case VKRRenderCommand::DRAW_INDEXED:
+					steps[i + 1]->commands.push_back(steps[j]->commands[k]);
+					break;
+				}
+			}
+			steps[j]->stepType = VKRStepType::RENDER_SKIP;
+		}
+
+		// We're done - we only expect one of these sequences per frame.
+		break;
+	}
 }
 
 void VulkanQueueRunner::ApplySonicHack(std::vector<VKRStep *> &steps) {