From 28b58ca05d188b81df2a247195fd35b54aef20a8 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Thu, 17 Sep 2020 00:49:17 +0200
Subject: [PATCH 1/8] More asserts

---
 Common/Vulkan/VulkanContext.cpp |  1 +
 Common/Vulkan/VulkanContext.h   | 31 ++++++++++++++++---------------
 GPU/Vulkan/DrawEngineVulkan.cpp |  9 ++++++---
 3 files changed, 23 insertions(+), 18 deletions(-)

diff --git a/Common/Vulkan/VulkanContext.cpp b/Common/Vulkan/VulkanContext.cpp
index 2cb47926f0..b659681a73 100644
--- a/Common/Vulkan/VulkanContext.cpp
+++ b/Common/Vulkan/VulkanContext.cpp
@@ -1,4 +1,5 @@
 #define __STDC_LIMIT_MACROS
+
 #include <cstdlib>
 #include <cstdint>
 #include <cstring>
diff --git a/Common/Vulkan/VulkanContext.h b/Common/Vulkan/VulkanContext.h
index 17509f6c72..fa12d26d79 100644
--- a/Common/Vulkan/VulkanContext.h
+++ b/Common/Vulkan/VulkanContext.h
@@ -5,6 +5,7 @@
 #include <vector>
 #include <utility>
 
+#include "Common/Log.h"
 #include "Common/Vulkan/VulkanLoader.h"
 #include "Common/Vulkan/VulkanDebug.h"
 
@@ -66,21 +67,21 @@ class VulkanDeleteList {
 
 public:
 	// NOTE: These all take reference handles so they can zero the input value.
-	void QueueDeleteCommandPool(VkCommandPool &pool) { cmdPools_.push_back(pool); pool = VK_NULL_HANDLE; }
-	void QueueDeleteDescriptorPool(VkDescriptorPool &pool) { descPools_.push_back(pool); pool = VK_NULL_HANDLE; }
-	void QueueDeleteShaderModule(VkShaderModule &module) { modules_.push_back(module); module = VK_NULL_HANDLE; }
-	void QueueDeleteBuffer(VkBuffer &buffer) { buffers_.push_back(buffer); buffer = VK_NULL_HANDLE; }
-	void QueueDeleteBufferView(VkBufferView &bufferView) { bufferViews_.push_back(bufferView); bufferView = VK_NULL_HANDLE; }
-	void QueueDeleteImage(VkImage &image) { images_.push_back(image); image = VK_NULL_HANDLE; }
-	void QueueDeleteImageView(VkImageView &imageView) { imageViews_.push_back(imageView); imageView = VK_NULL_HANDLE; }
-	void QueueDeleteDeviceMemory(VkDeviceMemory &deviceMemory) { deviceMemory_.push_back(deviceMemory); deviceMemory = VK_NULL_HANDLE; }
-	void QueueDeleteSampler(VkSampler &sampler) { samplers_.push_back(sampler); sampler = VK_NULL_HANDLE; }
-	void QueueDeletePipeline(VkPipeline &pipeline) { pipelines_.push_back(pipeline); pipeline = VK_NULL_HANDLE; }
-	void QueueDeletePipelineCache(VkPipelineCache &pipelineCache) { pipelineCaches_.push_back(pipelineCache); pipelineCache = VK_NULL_HANDLE; }
-	void QueueDeleteRenderPass(VkRenderPass &renderPass) { renderPasses_.push_back(renderPass); renderPass = VK_NULL_HANDLE; }
-	void QueueDeleteFramebuffer(VkFramebuffer &framebuffer) { framebuffers_.push_back(framebuffer); framebuffer = VK_NULL_HANDLE; }
-	void QueueDeletePipelineLayout(VkPipelineLayout &pipelineLayout) { pipelineLayouts_.push_back(pipelineLayout); pipelineLayout = VK_NULL_HANDLE; }
-	void QueueDeleteDescriptorSetLayout(VkDescriptorSetLayout &descSetLayout) { descSetLayouts_.push_back(descSetLayout); descSetLayout = VK_NULL_HANDLE; }
+	void QueueDeleteCommandPool(VkCommandPool &pool) { _dbg_assert_(pool != VK_NULL_HANDLE); cmdPools_.push_back(pool); pool = VK_NULL_HANDLE; }
+	void QueueDeleteDescriptorPool(VkDescriptorPool &pool) { _dbg_assert_(pool != VK_NULL_HANDLE); descPools_.push_back(pool); pool = VK_NULL_HANDLE; }
+	void QueueDeleteShaderModule(VkShaderModule &module) { _dbg_assert_(module != VK_NULL_HANDLE); modules_.push_back(module); module = VK_NULL_HANDLE; }
+	void QueueDeleteBuffer(VkBuffer &buffer) { _dbg_assert_(buffer != VK_NULL_HANDLE); buffers_.push_back(buffer); buffer = VK_NULL_HANDLE; }
+	void QueueDeleteBufferView(VkBufferView &bufferView) { _dbg_assert_(bufferView != VK_NULL_HANDLE); bufferViews_.push_back(bufferView); bufferView = VK_NULL_HANDLE; }
+	void QueueDeleteImage(VkImage &image) { _dbg_assert_(image != VK_NULL_HANDLE); images_.push_back(image); image = VK_NULL_HANDLE; }
+	void QueueDeleteImageView(VkImageView &imageView) { _dbg_assert_(imageView != VK_NULL_HANDLE); imageViews_.push_back(imageView); imageView = VK_NULL_HANDLE; }
+	void QueueDeleteDeviceMemory(VkDeviceMemory &deviceMemory) { _dbg_assert_(deviceMemory != VK_NULL_HANDLE); deviceMemory_.push_back(deviceMemory); deviceMemory = VK_NULL_HANDLE; }
+	void QueueDeleteSampler(VkSampler &sampler) { _dbg_assert_(sampler != VK_NULL_HANDLE); samplers_.push_back(sampler); sampler = VK_NULL_HANDLE; }
+	void QueueDeletePipeline(VkPipeline &pipeline) { _dbg_assert_(pipeline != VK_NULL_HANDLE); pipelines_.push_back(pipeline); pipeline = VK_NULL_HANDLE; }
+	void QueueDeletePipelineCache(VkPipelineCache &pipelineCache) { _dbg_assert_(pipelineCache != VK_NULL_HANDLE); pipelineCaches_.push_back(pipelineCache); pipelineCache = VK_NULL_HANDLE; }
+	void QueueDeleteRenderPass(VkRenderPass &renderPass) { _dbg_assert_(renderPass != VK_NULL_HANDLE); renderPasses_.push_back(renderPass); renderPass = VK_NULL_HANDLE; }
+	void QueueDeleteFramebuffer(VkFramebuffer &framebuffer) { _dbg_assert_(framebuffer != VK_NULL_HANDLE); framebuffers_.push_back(framebuffer); framebuffer = VK_NULL_HANDLE; }
+	void QueueDeletePipelineLayout(VkPipelineLayout &pipelineLayout) { _dbg_assert_(pipelineLayout != VK_NULL_HANDLE); pipelineLayouts_.push_back(pipelineLayout); pipelineLayout = VK_NULL_HANDLE; }
+	void QueueDeleteDescriptorSetLayout(VkDescriptorSetLayout &descSetLayout) { _dbg_assert_(descSetLayout != VK_NULL_HANDLE); descSetLayouts_.push_back(descSetLayout); descSetLayout = VK_NULL_HANDLE; }
 	void QueueCallback(void(*func)(void *userdata), void *userdata) { callbacks_.push_back(Callback(func, userdata)); }
 
 	void Take(VulkanDeleteList &del);
diff --git a/GPU/Vulkan/DrawEngineVulkan.cpp b/GPU/Vulkan/DrawEngineVulkan.cpp
index 4ac47d392e..f6c07886b3 100644
--- a/GPU/Vulkan/DrawEngineVulkan.cpp
+++ b/GPU/Vulkan/DrawEngineVulkan.cpp
@@ -388,6 +388,10 @@ VkResult DrawEngineVulkan::RecreateDescriptorPool(FrameData &frame, int newSize)
 }
 
 VkDescriptorSet DrawEngineVulkan::GetOrCreateDescriptorSet(VkImageView imageView, VkSampler sampler, VkBuffer base, VkBuffer light, VkBuffer bone, bool tess) {
+	_dbg_assert_(base != VK_NULL_HANDLE);
+	_dbg_assert_(light != VK_NULL_HANDLE);
+	_dbg_assert_(bone != VK_NULL_HANDLE);
+
 	DescriptorSetKey key;
 	key.imageView_ = imageView;
 	key.sampler_ = sampler;
@@ -396,9 +400,6 @@ VkDescriptorSet DrawEngineVulkan::GetOrCreateDescriptorSet(VkImageView imageView
 	key.base_ = base;
 	key.light_ = light;
 	key.bone_ = bone;
-	_dbg_assert_(base != VK_NULL_HANDLE);
-	_dbg_assert_(light != VK_NULL_HANDLE);
-	_dbg_assert_(bone != VK_NULL_HANDLE);
 
 	FrameData &frame = frame_[vulkan_->GetCurFrame()];
 	// See if we already have this descriptor set cached.
@@ -443,6 +444,8 @@ VkDescriptorSet DrawEngineVulkan::GetOrCreateDescriptorSet(VkImageView imageView
 	int n = 0;
 	VkDescriptorImageInfo tex[3]{};
 	if (imageView) {
+		_dbg_assert_(sampler != VK_NULL_HANDLE);
+
 #ifdef VULKAN_USE_GENERAL_LAYOUT_FOR_COLOR
 		tex[0].imageLayout = VK_IMAGE_LAYOUT_GENERAL;
 #else

From 608d082e49e041fb5a67b0aaf395e9875ab958d5 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Thu, 17 Sep 2020 18:54:17 +0200
Subject: [PATCH 2/8] FPURegCache: The invariant checking is very slow in debug
 mode. Let's only turn it on when we need it.

---
 Core/MIPS/x86/RegCacheFPU.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/Core/MIPS/x86/RegCacheFPU.cpp b/Core/MIPS/x86/RegCacheFPU.cpp
index 26e1e0dd55..e217bfc9e7 100644
--- a/Core/MIPS/x86/RegCacheFPU.cpp
+++ b/Core/MIPS/x86/RegCacheFPU.cpp
@@ -901,8 +901,8 @@ OpArg FPURegCache::GetDefaultLocation(int reg) const {
 }
 
 void FPURegCache::Invariant() const {
-#ifdef _DEBUG
-	_dbg_assert_msg_(SanityCheck() == 0, "Sanity check failed: %d", SanityCheck());
+#if 0
+	_assert_msg_(SanityCheck() == 0, "Sanity check failed: %d", SanityCheck());
 #endif
 }
 

From 181ddd842a4d6aeac3a1a190066ee7b6336db6fb Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Thu, 17 Sep 2020 19:18:32 +0200
Subject: [PATCH 3/8] Unify some GPU_SUPPORT* bits

---
 GPU/Common/FramebufferManagerCommon.cpp |  8 ++++++++
 GPU/D3D11/GPU_D3D11.cpp                 |  2 +-
 GPU/GLES/FramebufferManagerGLES.cpp     |  5 ++---
 GPU/GLES/GPU_GLES.cpp                   | 10 ++--------
 GPU/GLES/StencilBufferGLES.cpp          |  2 +-
 GPU/GPUState.h                          |  4 +---
 GPU/Vulkan/GPU_Vulkan.cpp               |  2 +-
 7 files changed, 16 insertions(+), 17 deletions(-)

diff --git a/GPU/Common/FramebufferManagerCommon.cpp b/GPU/Common/FramebufferManagerCommon.cpp
index 08972e868a..d26d635a86 100644
--- a/GPU/Common/FramebufferManagerCommon.cpp
+++ b/GPU/Common/FramebufferManagerCommon.cpp
@@ -347,6 +347,14 @@ VirtualFramebuffer *FramebufferManagerCommon::DoSetRenderFrameBuffer(const Frame
 		vfb->fb_stride = params.fb_stride;
 		vfb->z_address = params.z_address;
 		vfb->z_stride = params.z_stride;
+
+		if (vfb->z_address == vfb->fb_address) {
+			// Probably indicates that the game doesn't care about Z for this VFB.
+			// Let's avoid matching it for Z copies.
+			// vfb->z_address = 0;
+			// vfb->z_stride = 0;
+		}
+
 		vfb->width = drawing_width;
 		vfb->height = drawing_height;
 		vfb->newWidth = drawing_width;
diff --git a/GPU/D3D11/GPU_D3D11.cpp b/GPU/D3D11/GPU_D3D11.cpp
index 552d303555..cb053da7b0 100644
--- a/GPU/D3D11/GPU_D3D11.cpp
+++ b/GPU/D3D11/GPU_D3D11.cpp
@@ -150,7 +150,7 @@ void GPU_D3D11::CheckGPUFeatures() {
 	features |= GPU_SUPPORTS_TEXTURE_FLOAT;
 	features |= GPU_SUPPORTS_INSTANCE_RENDERING;
 	features |= GPU_SUPPORTS_TEXTURE_LOD_CONTROL;
-	features |= GPU_SUPPORTS_FBO;
+	features |= GPU_SUPPORTS_FRAMEBUFFER_BLIT;
 
 	uint32_t fmt4444 = draw_->GetDataFormatSupport(Draw::DataFormat::A4R4G4B4_UNORM_PACK16);
 	uint32_t fmt1555 = draw_->GetDataFormatSupport(Draw::DataFormat::A1R5G5B5_UNORM_PACK16);
diff --git a/GPU/GLES/FramebufferManagerGLES.cpp b/GPU/GLES/FramebufferManagerGLES.cpp
index 51bd76da3d..50c5fce743 100644
--- a/GPU/GLES/FramebufferManagerGLES.cpp
+++ b/GPU/GLES/FramebufferManagerGLES.cpp
@@ -272,7 +272,7 @@ void FramebufferManagerGLES::BlitFramebufferDepth(VirtualFramebuffer *src, Virtu
 		int w = std::min(src->renderWidth, dst->renderWidth);
 		int h = std::min(src->renderHeight, dst->renderHeight);
 
-		if (gstate_c.Supports(GPU_SUPPORTS_ARB_FRAMEBUFFER_BLIT | GPU_SUPPORTS_NV_FRAMEBUFFER_BLIT)) {
+		if (gstate_c.Supports(GPU_SUPPORTS_FRAMEBUFFER_BLIT)) {
 			// Let's only do this if not clearing depth.
 			draw_->BlitFramebuffer(src->fbo, 0, 0, w, h, dst->fbo, 0, 0, w, h, Draw::FB_DEPTH_BIT, Draw::FB_BLIT_NEAREST, "BlitFramebufferDepth");
 			dst->last_frame_depth_updated = gpuStats.numFlips;
@@ -330,8 +330,7 @@ void FramebufferManagerGLES::BlitFramebuffer(VirtualFramebuffer *dst, int dstX,
 		return;
 	}
 
-	bool useBlit = gstate_c.Supports(GPU_SUPPORTS_ARB_FRAMEBUFFER_BLIT | GPU_SUPPORTS_NV_FRAMEBUFFER_BLIT);
-	bool useNV = useBlit && !gstate_c.Supports(GPU_SUPPORTS_ARB_FRAMEBUFFER_BLIT);
+	bool useBlit = gstate_c.Supports(GPU_SUPPORTS_FRAMEBUFFER_BLIT);
 
 	float srcXFactor = useBlit ? (float)src->renderWidth / (float)src->bufferWidth : 1.0f;
 	float srcYFactor = useBlit ? (float)src->renderHeight / (float)src->bufferHeight : 1.0f;
diff --git a/GPU/GLES/GPU_GLES.cpp b/GPU/GLES/GPU_GLES.cpp
index e1de25d4e5..743682e369 100644
--- a/GPU/GLES/GPU_GLES.cpp
+++ b/GPU/GLES/GPU_GLES.cpp
@@ -182,14 +182,8 @@ void GPU_GLES::CheckGPUFeatures() {
 		}
 	}
 	
-	if (gl_extensions.ARB_framebuffer_object || gl_extensions.EXT_framebuffer_object || gl_extensions.IsGLES) {
-		features |= GPU_SUPPORTS_FBO;
-	}
-	if (gl_extensions.ARB_framebuffer_object || gl_extensions.GLES3) {
-		features |= GPU_SUPPORTS_ARB_FRAMEBUFFER_BLIT;
-	}
-	if (gl_extensions.NV_framebuffer_blit) {
-		features |= GPU_SUPPORTS_NV_FRAMEBUFFER_BLIT;
+	if (gl_extensions.ARB_framebuffer_object || gl_extensions.NV_framebuffer_blit || gl_extensions.GLES3) {
+		features |= GPU_SUPPORTS_FRAMEBUFFER_BLIT;
 	}
 	if (gl_extensions.ARB_vertex_array_object && gl_extensions.IsCoreContext) {
 		features |= GPU_SUPPORTS_VAO;
diff --git a/GPU/GLES/StencilBufferGLES.cpp b/GPU/GLES/StencilBufferGLES.cpp
index 8ca12e5c09..d9e7fadbf1 100644
--- a/GPU/GLES/StencilBufferGLES.cpp
+++ b/GPU/GLES/StencilBufferGLES.cpp
@@ -158,7 +158,7 @@ bool FramebufferManagerGLES::NotifyStencilUpload(u32 addr, int size, StencilUplo
 
 	shaderManagerGL_->DirtyLastShader();
 
-	bool useBlit = gstate_c.Supports(GPU_SUPPORTS_ARB_FRAMEBUFFER_BLIT | GPU_SUPPORTS_NV_FRAMEBUFFER_BLIT);
+	bool useBlit = gstate_c.Supports(GPU_SUPPORTS_FRAMEBUFFER_BLIT);
 
 	// Our fragment shader (and discard) is slow.  Since the source is 1x, we can stencil to 1x.
 	// Then after we're done, we'll just blit it across and stretch it there.
diff --git a/GPU/GPUState.h b/GPU/GPUState.h
index 2eac6bcef1..9ecf84bba2 100644
--- a/GPU/GPUState.h
+++ b/GPU/GPUState.h
@@ -489,9 +489,7 @@ enum {
 	GPU_ROUND_FRAGMENT_DEPTH_TO_16BIT = FLAG_BIT(22),
 	GPU_ROUND_DEPTH_TO_16BIT = FLAG_BIT(23),  // Can be disabled either per game or if we use a real 16-bit depth buffer
 	GPU_SUPPORTS_TEXTURE_LOD_CONTROL = FLAG_BIT(24),
-	GPU_SUPPORTS_FBO = FLAG_BIT(25),
-	GPU_SUPPORTS_ARB_FRAMEBUFFER_BLIT = FLAG_BIT(26),
-	GPU_SUPPORTS_NV_FRAMEBUFFER_BLIT = FLAG_BIT(27),
+	GPU_SUPPORTS_FRAMEBUFFER_BLIT = FLAG_BIT(26),
 	GPU_SUPPORTS_OES_TEXTURE_NPOT = FLAG_BIT(28),
 	GPU_NEEDS_Z_EQUAL_W_HACK = FLAG_BIT(29),
 	GPU_PREFER_CPU_DOWNLOAD = FLAG_BIT(30),
diff --git a/GPU/Vulkan/GPU_Vulkan.cpp b/GPU/Vulkan/GPU_Vulkan.cpp
index c24afc5f69..f6e63a7a30 100644
--- a/GPU/Vulkan/GPU_Vulkan.cpp
+++ b/GPU/Vulkan/GPU_Vulkan.cpp
@@ -224,7 +224,7 @@ void GPU_Vulkan::CheckGPUFeatures() {
 
 	// Mandatory features on Vulkan, which may be checked in "centralized" code
 	features |= GPU_SUPPORTS_TEXTURE_LOD_CONTROL;
-	features |= GPU_SUPPORTS_FBO;
+	features |= GPU_SUPPORTS_FRAMEBUFFER_BLIT;
 	features |= GPU_SUPPORTS_BLEND_MINMAX;
 	features |= GPU_SUPPORTS_ANY_COPY_IMAGE;
 	features |= GPU_SUPPORTS_OES_TEXTURE_NPOT;

From d0f7b1e30f58bf2611e8eaf657cb212a94b67528 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Thu, 17 Sep 2020 20:23:04 +0200
Subject: [PATCH 4/8] Make the implementations of CopyFramebufferDepth more
 similar.

Also avoid matching framebuffers with bad z for copies.
---
 GPU/Common/FramebufferManagerCommon.cpp |  4 ++--
 GPU/D3D11/FramebufferManagerD3D11.cpp   |  7 ++++---
 GPU/D3D11/GPU_D3D11.cpp                 |  2 +-
 GPU/Directx9/FramebufferManagerDX9.cpp  |  7 ++++++-
 GPU/GLES/FramebufferManagerGLES.cpp     | 14 ++++++--------
 GPU/GLES/GPU_GLES.cpp                   |  2 +-
 GPU/GPUState.h                          |  2 +-
 GPU/Vulkan/FramebufferManagerVulkan.cpp | 13 ++++---------
 GPU/Vulkan/GPU_Vulkan.cpp               |  2 +-
 ext/native/thin3d/thin3d_d3d9.cpp       |  4 +++-
 10 files changed, 29 insertions(+), 28 deletions(-)

diff --git a/GPU/Common/FramebufferManagerCommon.cpp b/GPU/Common/FramebufferManagerCommon.cpp
index d26d635a86..44162893bf 100644
--- a/GPU/Common/FramebufferManagerCommon.cpp
+++ b/GPU/Common/FramebufferManagerCommon.cpp
@@ -351,8 +351,8 @@ VirtualFramebuffer *FramebufferManagerCommon::DoSetRenderFrameBuffer(const Frame
 		if (vfb->z_address == vfb->fb_address) {
 			// Probably indicates that the game doesn't care about Z for this VFB.
 			// Let's avoid matching it for Z copies.
-			// vfb->z_address = 0;
-			// vfb->z_stride = 0;
+			vfb->z_address = 0;
+			vfb->z_stride = 0;
 		}
 
 		vfb->width = drawing_width;
diff --git a/GPU/D3D11/FramebufferManagerD3D11.cpp b/GPU/D3D11/FramebufferManagerD3D11.cpp
index 74ded8bc75..93648cbcbb 100644
--- a/GPU/D3D11/FramebufferManagerD3D11.cpp
+++ b/GPU/D3D11/FramebufferManagerD3D11.cpp
@@ -318,10 +318,11 @@ static void CopyPixelDepthOnly(u32 *dstp, const u32 *srcp, size_t c) {
 void FramebufferManagerD3D11::BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) {
 	bool matchingDepthBuffer = src->z_address == dst->z_address && src->z_stride != 0 && dst->z_stride != 0;
 	bool matchingSize = src->width == dst->width && src->height == dst->height;
-	bool matchingRenderSize = src->renderWidth == dst->renderWidth && src->renderHeight == dst->renderHeight;
-	if (matchingDepthBuffer && matchingSize && matchingRenderSize) {
+	if (matchingDepthBuffer && matchingSize) {
+		int w = std::min(src->renderWidth, dst->renderWidth);
+		int h = std::min(src->renderHeight, dst->renderHeight);
 		// TODO: Currently, this copies depth AND stencil, which is a problem.  See #9740.
-		draw_->CopyFramebufferImage(src->fbo, 0, 0, 0, 0, dst->fbo, 0, 0, 0, 0, src->renderWidth, src->renderHeight, 1, Draw::FB_DEPTH_BIT, "BlitFramebufferDepth");
+		draw_->CopyFramebufferImage(src->fbo, 0, 0, 0, 0, dst->fbo, 0, 0, 0, 0, w, h, 1, Draw::FB_DEPTH_BIT, "BlitFramebufferDepth");
 		RebindFramebuffer("RebindFramebuffer - BlitFramebufferDepth");
 		dst->last_frame_depth_updated = gpuStats.numFlips;
 	}
diff --git a/GPU/D3D11/GPU_D3D11.cpp b/GPU/D3D11/GPU_D3D11.cpp
index cb053da7b0..fcd579f5ed 100644
--- a/GPU/D3D11/GPU_D3D11.cpp
+++ b/GPU/D3D11/GPU_D3D11.cpp
@@ -146,7 +146,7 @@ void GPU_D3D11::CheckGPUFeatures() {
 		features |= GPU_SUPPORTS_DUALSOURCE_BLEND;
 	if (draw_->GetDeviceCaps().depthClampSupported)
 		features |= GPU_SUPPORTS_DEPTH_CLAMP;
-	features |= GPU_SUPPORTS_ANY_COPY_IMAGE;
+	features |= GPU_SUPPORTS_COPY_IMAGE;
 	features |= GPU_SUPPORTS_TEXTURE_FLOAT;
 	features |= GPU_SUPPORTS_INSTANCE_RENDERING;
 	features |= GPU_SUPPORTS_TEXTURE_LOD_CONTROL;
diff --git a/GPU/Directx9/FramebufferManagerDX9.cpp b/GPU/Directx9/FramebufferManagerDX9.cpp
index 012846f25c..3575ed82e9 100644
--- a/GPU/Directx9/FramebufferManagerDX9.cpp
+++ b/GPU/Directx9/FramebufferManagerDX9.cpp
@@ -309,7 +309,12 @@ static const D3DVERTEXELEMENT9 g_FramebufferVertexElements[] = {
 		bool matchingDepthBuffer = src->z_address == dst->z_address && src->z_stride != 0 && dst->z_stride != 0;
 		bool matchingSize = src->width == dst->width && src->height == dst->height;
 		if (matchingDepthBuffer && matchingSize) {
-			// Should use StretchRect here?  Note: should only copy depth and NOT copy stencil.  See #9740.
+			// Should use StretchRect here?  Note: should only copy depth and ideally NOT copy stencil.  See #9740.
+			int w = std::min(src->renderWidth, dst->renderWidth);
+			int h = std::min(src->renderHeight, dst->renderHeight);
+			draw_->CopyFramebufferImage(src->fbo, 0, 0, 0, 0, dst->fbo, 0, 0, 0, 0, w, h, 1, Draw::FB_DEPTH_BIT, "BlitFramebufferDepth");
+			RebindFramebuffer("RebindFramebuffer - BlitFramebufferDepth");
+			dst->last_frame_depth_updated = gpuStats.numFlips;
 		}
 	}
 
diff --git a/GPU/GLES/FramebufferManagerGLES.cpp b/GPU/GLES/FramebufferManagerGLES.cpp
index 50c5fce743..e030effb22 100644
--- a/GPU/GLES/FramebufferManagerGLES.cpp
+++ b/GPU/GLES/FramebufferManagerGLES.cpp
@@ -268,15 +268,13 @@ void FramebufferManagerGLES::BlitFramebufferDepth(VirtualFramebuffer *src, Virtu
 	bool matchingSize = src->width == dst->width && src->height == dst->height;
 
 	// Note: we don't use CopyFramebufferImage here, because it would copy depth AND stencil.  See #9740.
-	if (matchingDepthBuffer && matchingSize) {
+	if (matchingDepthBuffer && matchingSize && gstate_c.Supports(GPU_SUPPORTS_FRAMEBUFFER_BLIT)) {
 		int w = std::min(src->renderWidth, dst->renderWidth);
 		int h = std::min(src->renderHeight, dst->renderHeight);
-
-		if (gstate_c.Supports(GPU_SUPPORTS_FRAMEBUFFER_BLIT)) {
-			// Let's only do this if not clearing depth.
-			draw_->BlitFramebuffer(src->fbo, 0, 0, w, h, dst->fbo, 0, 0, w, h, Draw::FB_DEPTH_BIT, Draw::FB_BLIT_NEAREST, "BlitFramebufferDepth");
-			dst->last_frame_depth_updated = gpuStats.numFlips;
-		}
+		// Let's only do this if not clearing depth.
+		draw_->BlitFramebuffer(src->fbo, 0, 0, w, h, dst->fbo, 0, 0, w, h, Draw::FB_DEPTH_BIT, Draw::FB_BLIT_NEAREST, "BlitFramebufferDepth");
+		RebindFramebuffer("BlitFramebufferDepth");
+		dst->last_frame_depth_updated = gpuStats.numFlips;
 	}
 }
 
@@ -360,7 +358,7 @@ void FramebufferManagerGLES::BlitFramebuffer(VirtualFramebuffer *dst, int dstX,
 		return;
 	}
 
-	if (gstate_c.Supports(GPU_SUPPORTS_ANY_COPY_IMAGE)) {
+	if (gstate_c.Supports(GPU_SUPPORTS_COPY_IMAGE)) {
 		// glBlitFramebuffer can clip, but glCopyImageSubData is more restricted.
 		// In case the src goes outside, we just skip the optimization in that case.
 		const bool sameSize = dstX2 - dstX1 == srcX2 - srcX1 && dstY2 - dstY1 == srcY2 - srcY1;
diff --git a/GPU/GLES/GPU_GLES.cpp b/GPU/GLES/GPU_GLES.cpp
index 743682e369..2ba4922a2c 100644
--- a/GPU/GLES/GPU_GLES.cpp
+++ b/GPU/GLES/GPU_GLES.cpp
@@ -214,7 +214,7 @@ void GPU_GLES::CheckGPUFeatures() {
 		features |= GPU_SUPPORTS_BLEND_MINMAX;
 
 	if (gl_extensions.OES_copy_image || gl_extensions.NV_copy_image || gl_extensions.EXT_copy_image || gl_extensions.ARB_copy_image)
-		features |= GPU_SUPPORTS_ANY_COPY_IMAGE;
+		features |= GPU_SUPPORTS_COPY_IMAGE;
 
 	if (!gl_extensions.IsGLES)
 		features |= GPU_SUPPORTS_LOGIC_OP;
diff --git a/GPU/GPUState.h b/GPU/GPUState.h
index 9ecf84bba2..f012e023bc 100644
--- a/GPU/GPUState.h
+++ b/GPU/GPUState.h
@@ -483,7 +483,7 @@ enum {
 	// Free bit: 16
 	GPU_SUPPORTS_ACCURATE_DEPTH = FLAG_BIT(17),
 	GPU_SUPPORTS_VAO = FLAG_BIT(18),
-	GPU_SUPPORTS_ANY_COPY_IMAGE = FLAG_BIT(19),
+	GPU_SUPPORTS_COPY_IMAGE = FLAG_BIT(19),
 	GPU_SUPPORTS_ANY_FRAMEBUFFER_FETCH = FLAG_BIT(20),
 	GPU_SCALE_DEPTH_FROM_24BIT_TO_16BIT = FLAG_BIT(21),
 	GPU_ROUND_FRAGMENT_DEPTH_TO_16BIT = FLAG_BIT(22),
diff --git a/GPU/Vulkan/FramebufferManagerVulkan.cpp b/GPU/Vulkan/FramebufferManagerVulkan.cpp
index 9d57f8b85e..8a39fdc85e 100644
--- a/GPU/Vulkan/FramebufferManagerVulkan.cpp
+++ b/GPU/Vulkan/FramebufferManagerVulkan.cpp
@@ -273,17 +273,12 @@ void FramebufferManagerVulkan::ReformatFramebufferFrom(VirtualFramebuffer *vfb,
 void FramebufferManagerVulkan::BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) {
 	bool matchingDepthBuffer = src->z_address == dst->z_address && src->z_stride != 0 && dst->z_stride != 0;
 	bool matchingSize = src->width == dst->width && src->height == dst->height;
-	bool matchingRenderSize = src->renderWidth == dst->renderWidth && src->renderHeight == dst->renderHeight;
-	if (matchingDepthBuffer && matchingRenderSize && matchingSize) {
-		// TODO: Currently, this copies depth AND stencil, which is a problem.  See #9740.
-		draw_->CopyFramebufferImage(src->fbo, 0, 0, 0, 0, dst->fbo, 0, 0, 0, 0, src->renderWidth, src->renderHeight, 1, Draw::FB_DEPTH_BIT, "BlitFramebufferDepth");
-		dst->last_frame_depth_updated = gpuStats.numFlips;
-	} else if (matchingDepthBuffer && matchingSize) {
-		/*
+	if (matchingDepthBuffer && matchingSize) {
 		int w = std::min(src->renderWidth, dst->renderWidth);
 		int h = std::min(src->renderHeight, dst->renderHeight);
-		draw_->BlitFramebuffer(src->fbo, 0, 0, w, h, dst->fbo, 0, 0, w, h, Draw::FB_DEPTH_BIT, Draw::FB_BLIT_NEAREST);
-		*/
+		// TODO: Currently, this copies depth AND stencil, which is a problem.  See #9740.
+		draw_->CopyFramebufferImage(src->fbo, 0, 0, 0, 0, dst->fbo, 0, 0, 0, 0, w, h, 1, Draw::FB_DEPTH_BIT, "BlitFramebufferDepth");
+		dst->last_frame_depth_updated = gpuStats.numFlips;
 	}
 }
 
diff --git a/GPU/Vulkan/GPU_Vulkan.cpp b/GPU/Vulkan/GPU_Vulkan.cpp
index f6e63a7a30..9c7f36afe7 100644
--- a/GPU/Vulkan/GPU_Vulkan.cpp
+++ b/GPU/Vulkan/GPU_Vulkan.cpp
@@ -226,7 +226,7 @@ void GPU_Vulkan::CheckGPUFeatures() {
 	features |= GPU_SUPPORTS_TEXTURE_LOD_CONTROL;
 	features |= GPU_SUPPORTS_FRAMEBUFFER_BLIT;
 	features |= GPU_SUPPORTS_BLEND_MINMAX;
-	features |= GPU_SUPPORTS_ANY_COPY_IMAGE;
+	features |= GPU_SUPPORTS_COPY_IMAGE;
 	features |= GPU_SUPPORTS_OES_TEXTURE_NPOT;
 	features |= GPU_SUPPORTS_INSTANCE_RENDERING;
 	features |= GPU_SUPPORTS_VERTEX_TEXTURE_FETCH;
diff --git a/ext/native/thin3d/thin3d_d3d9.cpp b/ext/native/thin3d/thin3d_d3d9.cpp
index 9843af9358..8a97cbd26e 100644
--- a/ext/native/thin3d/thin3d_d3d9.cpp
+++ b/ext/native/thin3d/thin3d_d3d9.cpp
@@ -526,7 +526,9 @@ public:
 
 	void UpdateBuffer(Buffer *buffer, const uint8_t *data, size_t offset, size_t size, UpdateBufferFlags flags) override;
 
-	void CopyFramebufferImage(Framebuffer *src, int level, int x, int y, int z, Framebuffer *dst, int dstLevel, int dstX, int dstY, int dstZ, int width, int height, int depth, int channelBits, const char *tag) override {}
+	void CopyFramebufferImage(Framebuffer *src, int level, int x, int y, int z, Framebuffer *dst, int dstLevel, int dstX, int dstY, int dstZ, int width, int height, int depth, int channelBits, const char *tag) override {
+		// Not implemented
+	}
 	bool BlitFramebuffer(Framebuffer *src, int srcX1, int srcY1, int srcX2, int srcY2, Framebuffer *dst, int dstX1, int dstY1, int dstX2, int dstY2, int channelBits, FBBlitFilter filter, const char *tag) override;
 
 	// These functions should be self explanatory.

From d6d72db0a9717b8eefa874a0fc9806bf83a00c7f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Thu, 17 Sep 2020 20:31:40 +0200
Subject: [PATCH 5/8] Unify BlitFramebufferDepth.

---
 GPU/Common/FramebufferManagerCommon.cpp | 20 ++++++++++++++++++++
 GPU/Common/FramebufferManagerCommon.h   |  2 +-
 GPU/D3D11/FramebufferManagerD3D11.cpp   | 13 -------------
 GPU/D3D11/FramebufferManagerD3D11.h     |  2 --
 GPU/Directx9/FramebufferManagerDX9.cpp  | 13 -------------
 GPU/Directx9/FramebufferManagerDX9.h    |  2 --
 GPU/GLES/FramebufferManagerGLES.cpp     | 15 ---------------
 GPU/GLES/FramebufferManagerGLES.h       |  2 --
 GPU/Vulkan/FramebufferManagerVulkan.cpp | 13 -------------
 GPU/Vulkan/FramebufferManagerVulkan.h   |  2 --
 10 files changed, 21 insertions(+), 63 deletions(-)

diff --git a/GPU/Common/FramebufferManagerCommon.cpp b/GPU/Common/FramebufferManagerCommon.cpp
index 44162893bf..5776bc4919 100644
--- a/GPU/Common/FramebufferManagerCommon.cpp
+++ b/GPU/Common/FramebufferManagerCommon.cpp
@@ -469,6 +469,26 @@ void FramebufferManagerCommon::DestroyFramebuf(VirtualFramebuffer *v) {
 	delete v;
 }
 
+void FramebufferManagerCommon::BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) {
+	bool matchingDepthBuffer = src->z_address == dst->z_address && src->z_stride != 0 && dst->z_stride != 0;
+	bool matchingSize = src->width == dst->width && src->height == dst->height;
+
+	// Note: we don't use CopyFramebufferImage here, because it would copy depth AND stencil.  See #9740.
+	if (matchingDepthBuffer && matchingSize) {
+		int w = std::min(src->renderWidth, dst->renderWidth);
+		int h = std::min(src->renderHeight, dst->renderHeight);
+		// Let's only do this if not clearing depth.
+		if (gstate_c.Supports(GPU_SUPPORTS_FRAMEBUFFER_BLIT)) {
+			draw_->BlitFramebuffer(src->fbo, 0, 0, w, h, dst->fbo, 0, 0, w, h, Draw::FB_DEPTH_BIT, Draw::FB_BLIT_NEAREST, "BlitFramebufferDepth");
+			RebindFramebuffer("BlitFramebufferDepth");
+		} else if (gstate_c.Supports(GPU_SUPPORTS_COPY_IMAGE)) {
+			draw_->CopyFramebufferImage(src->fbo, 0, 0, 0, 0, dst->fbo, 0, 0, 0, 0, w, h, 1, Draw::FB_DEPTH_BIT, "BlitFramebufferDepth");
+			RebindFramebuffer("BlitFramebufferDepth");
+		}
+		dst->last_frame_depth_updated = gpuStats.numFlips;
+	}
+}
+
 void FramebufferManagerCommon::NotifyRenderFramebufferCreated(VirtualFramebuffer *vfb) {
 	if (!useBufferedRendering_) {
 		// Let's ignore rendering to targets that have not (yet) been displayed.
diff --git a/GPU/Common/FramebufferManagerCommon.h b/GPU/Common/FramebufferManagerCommon.h
index ff43f6d504..05a2a787b6 100644
--- a/GPU/Common/FramebufferManagerCommon.h
+++ b/GPU/Common/FramebufferManagerCommon.h
@@ -339,7 +339,7 @@ protected:
 	void NotifyRenderFramebufferSwitched(VirtualFramebuffer *prevVfb, VirtualFramebuffer *vfb, bool isClearingDepth);
 
 	virtual void ReformatFramebufferFrom(VirtualFramebuffer *vfb, GEBufferFormat old) = 0;
-	virtual void BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) = 0;
+	void BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst);
 
 	void ResizeFramebufFBO(VirtualFramebuffer *vfb, int w, int h, bool force = false, bool skipCopy = false);
 	void ShowScreenResolution();
diff --git a/GPU/D3D11/FramebufferManagerD3D11.cpp b/GPU/D3D11/FramebufferManagerD3D11.cpp
index 93648cbcbb..4440ebf4b3 100644
--- a/GPU/D3D11/FramebufferManagerD3D11.cpp
+++ b/GPU/D3D11/FramebufferManagerD3D11.cpp
@@ -315,19 +315,6 @@ static void CopyPixelDepthOnly(u32 *dstp, const u32 *srcp, size_t c) {
 	}
 }
 
-void FramebufferManagerD3D11::BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) {
-	bool matchingDepthBuffer = src->z_address == dst->z_address && src->z_stride != 0 && dst->z_stride != 0;
-	bool matchingSize = src->width == dst->width && src->height == dst->height;
-	if (matchingDepthBuffer && matchingSize) {
-		int w = std::min(src->renderWidth, dst->renderWidth);
-		int h = std::min(src->renderHeight, dst->renderHeight);
-		// TODO: Currently, this copies depth AND stencil, which is a problem.  See #9740.
-		draw_->CopyFramebufferImage(src->fbo, 0, 0, 0, 0, dst->fbo, 0, 0, 0, 0, w, h, 1, Draw::FB_DEPTH_BIT, "BlitFramebufferDepth");
-		RebindFramebuffer("RebindFramebuffer - BlitFramebufferDepth");
-		dst->last_frame_depth_updated = gpuStats.numFlips;
-	}
-}
-
 void FramebufferManagerD3D11::BindFramebufferAsColorTexture(int stage, VirtualFramebuffer *framebuffer, int flags) {
 	if (!framebuffer->fbo || !useBufferedRendering_) {
 		ID3D11ShaderResourceView *view = nullptr;
diff --git a/GPU/D3D11/FramebufferManagerD3D11.h b/GPU/D3D11/FramebufferManagerD3D11.h
index a1683c299c..e951e277a1 100644
--- a/GPU/D3D11/FramebufferManagerD3D11.h
+++ b/GPU/D3D11/FramebufferManagerD3D11.h
@@ -45,8 +45,6 @@ public:
 	void DeviceLost();
 	void ReformatFramebufferFrom(VirtualFramebuffer *vfb, GEBufferFormat old) override;
 
-	void BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) override;
-
 	void BindFramebufferAsColorTexture(int stage, VirtualFramebuffer *framebuffer, int flags);
 
 	virtual bool NotifyStencilUpload(u32 addr, int size, StencilUpload flags = StencilUpload::NEEDS_CLEAR) override;
diff --git a/GPU/Directx9/FramebufferManagerDX9.cpp b/GPU/Directx9/FramebufferManagerDX9.cpp
index 3575ed82e9..be7f751eda 100644
--- a/GPU/Directx9/FramebufferManagerDX9.cpp
+++ b/GPU/Directx9/FramebufferManagerDX9.cpp
@@ -305,19 +305,6 @@ static const D3DVERTEXELEMENT9 g_FramebufferVertexElements[] = {
 		}
 	}
 
-	void FramebufferManagerDX9::BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) {
-		bool matchingDepthBuffer = src->z_address == dst->z_address && src->z_stride != 0 && dst->z_stride != 0;
-		bool matchingSize = src->width == dst->width && src->height == dst->height;
-		if (matchingDepthBuffer && matchingSize) {
-			// Should use StretchRect here?  Note: should only copy depth and ideally NOT copy stencil.  See #9740.
-			int w = std::min(src->renderWidth, dst->renderWidth);
-			int h = std::min(src->renderHeight, dst->renderHeight);
-			draw_->CopyFramebufferImage(src->fbo, 0, 0, 0, 0, dst->fbo, 0, 0, 0, 0, w, h, 1, Draw::FB_DEPTH_BIT, "BlitFramebufferDepth");
-			RebindFramebuffer("RebindFramebuffer - BlitFramebufferDepth");
-			dst->last_frame_depth_updated = gpuStats.numFlips;
-		}
-	}
-
 	LPDIRECT3DSURFACE9 FramebufferManagerDX9::GetOffscreenSurface(LPDIRECT3DSURFACE9 similarSurface, VirtualFramebuffer *vfb) {
 		D3DSURFACE_DESC desc = {};
 		HRESULT hr = similarSurface->GetDesc(&desc);
diff --git a/GPU/Directx9/FramebufferManagerDX9.h b/GPU/Directx9/FramebufferManagerDX9.h
index b164a9b0e7..3a8578599a 100644
--- a/GPU/Directx9/FramebufferManagerDX9.h
+++ b/GPU/Directx9/FramebufferManagerDX9.h
@@ -50,8 +50,6 @@ public:
 	void DeviceLost();
 	void ReformatFramebufferFrom(VirtualFramebuffer *vfb, GEBufferFormat old) override;
 
-	void BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) override;
-
 	void BindFramebufferAsColorTexture(int stage, VirtualFramebuffer *framebuffer, int flags);
 
 	virtual bool NotifyStencilUpload(u32 addr, int size, StencilUpload flags = StencilUpload::NEEDS_CLEAR) override;
diff --git a/GPU/GLES/FramebufferManagerGLES.cpp b/GPU/GLES/FramebufferManagerGLES.cpp
index e030effb22..f2709663cf 100644
--- a/GPU/GLES/FramebufferManagerGLES.cpp
+++ b/GPU/GLES/FramebufferManagerGLES.cpp
@@ -263,21 +263,6 @@ void FramebufferManagerGLES::ReformatFramebufferFrom(VirtualFramebuffer *vfb, GE
 	}
 }
 
-void FramebufferManagerGLES::BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) {
-	bool matchingDepthBuffer = src->z_address == dst->z_address && src->z_stride != 0 && dst->z_stride != 0;
-	bool matchingSize = src->width == dst->width && src->height == dst->height;
-
-	// Note: we don't use CopyFramebufferImage here, because it would copy depth AND stencil.  See #9740.
-	if (matchingDepthBuffer && matchingSize && gstate_c.Supports(GPU_SUPPORTS_FRAMEBUFFER_BLIT)) {
-		int w = std::min(src->renderWidth, dst->renderWidth);
-		int h = std::min(src->renderHeight, dst->renderHeight);
-		// Let's only do this if not clearing depth.
-		draw_->BlitFramebuffer(src->fbo, 0, 0, w, h, dst->fbo, 0, 0, w, h, Draw::FB_DEPTH_BIT, Draw::FB_BLIT_NEAREST, "BlitFramebufferDepth");
-		RebindFramebuffer("BlitFramebufferDepth");
-		dst->last_frame_depth_updated = gpuStats.numFlips;
-	}
-}
-
 void FramebufferManagerGLES::BindFramebufferAsColorTexture(int stage, VirtualFramebuffer *framebuffer, int flags) {
 	if (!framebuffer->fbo || !useBufferedRendering_) {
 		render_->BindTexture(stage, nullptr);
diff --git a/GPU/GLES/FramebufferManagerGLES.h b/GPU/GLES/FramebufferManagerGLES.h
index 12a3eb85c5..9177b62767 100644
--- a/GPU/GLES/FramebufferManagerGLES.h
+++ b/GPU/GLES/FramebufferManagerGLES.h
@@ -49,8 +49,6 @@ public:
 	void DeviceLost();
 	void ReformatFramebufferFrom(VirtualFramebuffer *vfb, GEBufferFormat old) override;
 
-	void BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) override;
-
 	// For use when texturing from a framebuffer.  May create a duplicate if target.
 	void BindFramebufferAsColorTexture(int stage, VirtualFramebuffer *framebuffer, int flags);
 
diff --git a/GPU/Vulkan/FramebufferManagerVulkan.cpp b/GPU/Vulkan/FramebufferManagerVulkan.cpp
index 8a39fdc85e..293ff77be5 100644
--- a/GPU/Vulkan/FramebufferManagerVulkan.cpp
+++ b/GPU/Vulkan/FramebufferManagerVulkan.cpp
@@ -269,19 +269,6 @@ void FramebufferManagerVulkan::ReformatFramebufferFrom(VirtualFramebuffer *vfb,
 	}
 }
 
-// Except for a missing rebind and silly scissor enables, identical copy of the same function in GPU_GLES - tricky parts are in thin3d.
-void FramebufferManagerVulkan::BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) {
-	bool matchingDepthBuffer = src->z_address == dst->z_address && src->z_stride != 0 && dst->z_stride != 0;
-	bool matchingSize = src->width == dst->width && src->height == dst->height;
-	if (matchingDepthBuffer && matchingSize) {
-		int w = std::min(src->renderWidth, dst->renderWidth);
-		int h = std::min(src->renderHeight, dst->renderHeight);
-		// TODO: Currently, this copies depth AND stencil, which is a problem.  See #9740.
-		draw_->CopyFramebufferImage(src->fbo, 0, 0, 0, 0, dst->fbo, 0, 0, 0, 0, w, h, 1, Draw::FB_DEPTH_BIT, "BlitFramebufferDepth");
-		dst->last_frame_depth_updated = gpuStats.numFlips;
-	}
-}
-
 VkImageView FramebufferManagerVulkan::BindFramebufferAsColorTexture(int stage, VirtualFramebuffer *framebuffer, int flags) {
 	if (!framebuffer->fbo || !useBufferedRendering_) {
 		gstate_c.skipDrawReason |= SKIPDRAW_BAD_FB_TEXTURE;
diff --git a/GPU/Vulkan/FramebufferManagerVulkan.h b/GPU/Vulkan/FramebufferManagerVulkan.h
index b66eeb6033..a09f6ef755 100644
--- a/GPU/Vulkan/FramebufferManagerVulkan.h
+++ b/GPU/Vulkan/FramebufferManagerVulkan.h
@@ -54,8 +54,6 @@ public:
 	int GetLineWidth();
 	void ReformatFramebufferFrom(VirtualFramebuffer *vfb, GEBufferFormat old) override;
 
-	void BlitFramebufferDepth(VirtualFramebuffer *src, VirtualFramebuffer *dst) override;
-
 	bool NotifyStencilUpload(u32 addr, int size, StencilUpload flags = StencilUpload::NEEDS_CLEAR) override;
 
 	VkImageView BindFramebufferAsColorTexture(int stage, VirtualFramebuffer *framebuffer, int flags);

From d71ec8d82c8a60c9dc6b6389718eda98b3823840 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Thu, 17 Sep 2020 20:33:19 +0200
Subject: [PATCH 6/8] The new asserts found something.

---
 ext/native/thin3d/VulkanQueueRunner.cpp | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/ext/native/thin3d/VulkanQueueRunner.cpp b/ext/native/thin3d/VulkanQueueRunner.cpp
index b97a9a0f4a..991374fb92 100644
--- a/ext/native/thin3d/VulkanQueueRunner.cpp
+++ b/ext/native/thin3d/VulkanQueueRunner.cpp
@@ -89,8 +89,12 @@ void VulkanQueueRunner::ResizeReadbackBuffer(VkDeviceSize requiredSize) {
 
 void VulkanQueueRunner::DestroyDeviceObjects() {
 	INFO_LOG(G3D, "VulkanQueueRunner::DestroyDeviceObjects");
-	vulkan_->Delete().QueueDeleteDeviceMemory(readbackMemory_);
-	vulkan_->Delete().QueueDeleteBuffer(readbackBuffer_);
+	if (readbackMemory_) {
+		vulkan_->Delete().QueueDeleteDeviceMemory(readbackMemory_);
+	}
+	if (readbackBuffer_) {
+		vulkan_->Delete().QueueDeleteBuffer(readbackBuffer_);
+	}
 	readbackBufferSize_ = 0;
 
 	renderPasses_.Iterate([&](const RPKey &rpkey, VkRenderPass rp) {

From 277f0e4ec6fd08523f0911808c025c37bca14690 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Thu, 17 Sep 2020 20:47:22 +0200
Subject: [PATCH 7/8] Fix bug in GLES texture filter setup

---
 GPU/GLES/TextureCacheGLES.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/GPU/GLES/TextureCacheGLES.cpp b/GPU/GLES/TextureCacheGLES.cpp
index 292fa20ba1..de6852d556 100644
--- a/GPU/GLES/TextureCacheGLES.cpp
+++ b/GPU/GLES/TextureCacheGLES.cpp
@@ -124,10 +124,10 @@ void TextureCacheGLES::ApplySamplingParams(const SamplerCacheKey &key) {
 	}
 
 	float aniso = 0.0f;
-	int magKey = ((int)key.mipEnable << 2) | ((int)key.mipFilt << 1) | ((int)key.magFilt);
+	int minKey = ((int)key.mipEnable << 2) | ((int)key.mipFilt << 1) | ((int)key.minFilt);
 	render_->SetTextureSampler(0,
 		key.sClamp ? GL_CLAMP_TO_EDGE : GL_REPEAT, key.tClamp ? GL_CLAMP_TO_EDGE : GL_REPEAT,
-		MagFiltGL[magKey], key.minFilt ? GL_LINEAR : GL_NEAREST, aniso);
+		key.magFilt ? GL_LINEAR : GL_NEAREST, MinFiltGL[minKey], aniso);
 }
 
 static void ConvertColors(void *dstBuf, const void *srcBuf, Draw::DataFormat dstFmt, int numPixels) {

From 7c76f4e65ca3dd9a27dc8033cd651253c3eb8e2e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Henrik=20Rydg=C3=A5rd?= <hrydgard@gmail.com>
Date: Thu, 17 Sep 2020 20:48:24 +0200
Subject: [PATCH 8/8] Minor improvement to gl_debug_log.h

---
 ext/native/gfx/gl_debug_log.cpp     | 4 +++-
 ext/native/gfx/gl_debug_log.h       | 4 ++--
 ext/native/thin3d/GLQueueRunner.cpp | 4 ++++
 3 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/ext/native/gfx/gl_debug_log.cpp b/ext/native/gfx/gl_debug_log.cpp
index 078b4ed843..fc66ccf1dd 100644
--- a/ext/native/gfx/gl_debug_log.cpp
+++ b/ext/native/gfx/gl_debug_log.cpp
@@ -34,9 +34,11 @@ std::string GLEnumToString(uint16_t value) {
 	}
 }
 
-void CheckGLError(const char *file, int line) {
+bool CheckGLError(const char *file, int line) {
 	GLenum err = glGetError();
 	if (err != GL_NO_ERROR) {
 		ERROR_LOG(G3D, "GL error %s on %s:%d", GLEnumToString(err).c_str(), file, line);
+		return false;
 	}
+	return true;
 }
diff --git a/ext/native/gfx/gl_debug_log.h b/ext/native/gfx/gl_debug_log.h
index ae22ab1700..d6d37c4271 100644
--- a/ext/native/gfx/gl_debug_log.h
+++ b/ext/native/gfx/gl_debug_log.h
@@ -10,8 +10,8 @@
 
 #if defined(DEBUG_OPENGL)
 
-void CheckGLError(const char *file, int line);
-#define CHECK_GL_ERROR_IF_DEBUG() CheckGLError(__FILE__, __LINE__)
+bool CheckGLError(const char *file, int line);
+#define CHECK_GL_ERROR_IF_DEBUG() if (!CheckGLError(__FILE__, __LINE__)) __debugbreak();
 
 #else
 
diff --git a/ext/native/thin3d/GLQueueRunner.cpp b/ext/native/thin3d/GLQueueRunner.cpp
index 9c380ee622..c5da0e4c21 100644
--- a/ext/native/thin3d/GLQueueRunner.cpp
+++ b/ext/native/thin3d/GLQueueRunner.cpp
@@ -1146,6 +1146,7 @@ void GLQueueRunner::PerformRenderPass(const GLRStep &step, bool first, bool last
 			if (!tex) {
 				break;
 			}
+			CHECK_GL_ERROR_IF_DEBUG();
 			if (tex->canWrap) {
 				if (tex->wrapS != c.textureSampler.wrapS) {
 					glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, c.textureSampler.wrapS);
@@ -1156,14 +1157,17 @@ void GLQueueRunner::PerformRenderPass(const GLRStep &step, bool first, bool last
 					tex->wrapT = c.textureSampler.wrapT;
 				}
 			}
+			CHECK_GL_ERROR_IF_DEBUG();
 			if (tex->magFilter != c.textureSampler.magFilter) {
 				glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, c.textureSampler.magFilter);
 				tex->magFilter = c.textureSampler.magFilter;
 			}
+			CHECK_GL_ERROR_IF_DEBUG();
 			if (tex->minFilter != c.textureSampler.minFilter) {
 				glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, c.textureSampler.minFilter);
 				tex->minFilter = c.textureSampler.minFilter;
 			}
+			CHECK_GL_ERROR_IF_DEBUG();
 			if (tex->anisotropy != c.textureSampler.anisotropy) {
 				if (c.textureSampler.anisotropy != 0.0f) {
 					glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAX_ANISOTROPY_EXT, c.textureSampler.anisotropy);