From df8dac4aae74e52aae1eeb3d7ede5ce81fbffa03 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Fri, 16 Mar 2018 13:35:49 +0100
Subject: [PATCH] If the first renderpass is an empty clear, merge it into the
 next one that touches the same framebuffer. Works around #10723 in Wipeout.

---
 ext/native/thin3d/VulkanQueueRunner.cpp | 37 +++++++++++++++++++++++--
 ext/native/thin3d/VulkanQueueRunner.h   |  1 +
 2 files changed, 36 insertions(+), 2 deletions(-)

diff --git a/ext/native/thin3d/VulkanQueueRunner.cpp b/ext/native/thin3d/VulkanQueueRunner.cpp
index 235000821f..fdeafcd6b2 100644
--- a/ext/native/thin3d/VulkanQueueRunner.cpp
+++ b/ext/native/thin3d/VulkanQueueRunner.cpp
@@ -255,11 +255,39 @@ VkRenderPass VulkanQueueRunner::GetRenderPass(VKRRenderPassAction colorLoadActio
 void VulkanQueueRunner::RunSteps(VkCommandBuffer cmd, const std::vector<VKRStep *> &steps) {
 	// Optimizes renderpasses, then sequences them.
 	// Planned optimizations: 
-	//  * Push down empty "Clear/Store" renderpasses, and merge them with the first "Load/Store" to the same framebuffer.
-	//    (These happen in Wipeout)
 	//  * Create copies of render target that are rendered to multiple times and textured from in sequence, and push those render passes
 	//    as early as possible in the frame (Wipeout billboards).
 
+	// Push down empty "Clear/Store" renderpasses, and merge them with the first "Load/Store" to the same framebuffer.
+	// Actually let's just bother with the first one for now. This affects Wipeout Pure.
+	if (steps.size() > 1 && steps[0]->stepType == VKRStepType::RENDER &&
+		steps[0]->render.numDraws == 0 &&
+		steps[0]->render.color == VKRRenderPassAction::CLEAR &&
+		steps[0]->render.stencil == VKRRenderPassAction::CLEAR &&
+		steps[0]->render.depth == VKRRenderPassAction::CLEAR) {
+		// Drop the first step, and merge it into the next step that touches the same framebuffer.
+		for (size_t i = 1; i < steps.size(); i++) {
+			if (steps[i]->stepType == VKRStepType::RENDER &&
+				steps[i]->render.framebuffer == steps[0]->render.framebuffer) {
+				if (steps[i]->render.color != VKRRenderPassAction::CLEAR) {
+					steps[i]->render.color = VKRRenderPassAction::CLEAR;
+					steps[i]->render.clearColor = steps[0]->render.clearColor;
+				}
+				if (steps[i]->render.depth != VKRRenderPassAction::CLEAR) {
+					steps[i]->render.depth = VKRRenderPassAction::CLEAR;
+					steps[i]->render.clearDepth = steps[0]->render.clearDepth;
+				}
+				if (steps[i]->render.stencil != VKRRenderPassAction::CLEAR) {
+					steps[i]->render.stencil = VKRRenderPassAction::CLEAR;
+					steps[i]->render.clearStencil = steps[0]->render.clearStencil;
+				}
+				// Cheaply skip the first step.
+				steps[0]->stepType = VKRStepType::RENDER_SKIP;
+				break;
+			}
+		}
+	}
+
 	for (size_t i = 0; i < steps.size(); i++) {
 		const VKRStep &step = *steps[i];
 		switch (step.stepType) {
@@ -278,6 +306,8 @@ void VulkanQueueRunner::RunSteps(VkCommandBuffer cmd, const std::vector<VKRStep
 		case VKRStepType::READBACK_IMAGE:
 			PerformReadbackImage(step, cmd);
 			break;
+		case VKRStepType::RENDER_SKIP:
+			break;
 		}
 		delete steps[i];
 	}
@@ -303,6 +333,9 @@ void VulkanQueueRunner::LogSteps(const std::vector<VKRStep *> &steps) {
 		case VKRStepType::READBACK_IMAGE:
 			LogReadbackImage(step);
 			break;
+		case VKRStepType::RENDER_SKIP:
+			ILOG("(skipped render pass)");
+			break;
 		}
 	}
 }
diff --git a/ext/native/thin3d/VulkanQueueRunner.h b/ext/native/thin3d/VulkanQueueRunner.h
index 8257fa450a..f10326ff95 100644
--- a/ext/native/thin3d/VulkanQueueRunner.h
+++ b/ext/native/thin3d/VulkanQueueRunner.h
@@ -82,6 +82,7 @@ struct VkRenderData {
 
 enum class VKRStepType : uint8_t {
 	RENDER,
+	RENDER_SKIP,
 	COPY,
 	BLIT,
 	READBACK,