VIF: Fix out of bounds read caused by broken alignment logic

Ivy Bridge and below default to Direct3D11.

Haswell/Broadwell default to OpenGL.

Skylake and above default to Direct3D12.

pcsx2/GS/Renderers/Common/GSTexture.cpp

@@ -200,7 +200,7 @@ u32 GSTexture::CalcUploadSize(Format format, u32 height, u32 pitch)
 
 bool GSTexture::IsFeedbackFormat(Format format)
 {
-	return format == Format::Color || format == Format::ColorClip ||
+	return format == Format::Color || format == Format::ColorClip || format == Format::ColorHDR ||
 		format == Format::DepthColor || format == Format::DepthStencil;
 }
 

pcsx2/GS/Renderers/DX11/D3D.cpp

@@ -374,6 +374,8 @@ GSRendererType D3D::GetPreferredRenderer()
 		Console.WriteLn("D3D11 feature level for autodetection: %x", static_cast<unsigned>(feature_level));
 		return feature_level;
 	};
+
+	/*
 	const auto get_d3d12_device = [&adapter]() {
 		wil::com_ptr_nothrow<ID3D12Device> device;
 		const HRESULT hr = D3D12CreateDevice(adapter.get(), D3D_FEATURE_LEVEL_12_0, IID_PPV_ARGS(device.put()));
@@ -381,6 +383,7 @@ GSRendererType D3D::GetPreferredRenderer()
 			Console.Error("D3D12CreateDevice() for automatic renderer failed: %08X", hr);
 		return device;
 	};
+	
 #ifdef ENABLE_VULKAN
 	static constexpr auto check_vulkan_supported = []() {
 		if (!GSDeviceVK::EnumerateGPUs().empty())
@@ -395,6 +398,7 @@ GSRendererType D3D::GetPreferredRenderer()
 #else
 	static constexpr auto check_vulkan_supported = []() { return false; };
 #endif
+	*/
 
 	switch (GetVendorID(adapter.get()))
 	{
@@ -432,26 +436,37 @@ GSRendererType D3D::GetPreferredRenderer()
 
 			// Sampler feedback Tier 0.9 is only present in Tiger Lake/Xe/Arc, so we can use that to
 			// differentiate between them. Unfortunately, that requires a D3D12 device.
-			const auto device12 = get_d3d12_device();
-			if (device12)
+			const std::optional<D3D_FEATURE_LEVEL> feature_level = get_d3d11_feature_level();
+			if (!feature_level.has_value())
+				return GSRendererType::DX11;
+			else if (feature_level == D3D_FEATURE_LEVEL_12_0)
 			{
-				D3D12_FEATURE_DATA_D3D12_OPTIONS7 opts = {};
-				if (SUCCEEDED(device12->CheckFeatureSupport(D3D12_FEATURE_D3D12_OPTIONS7, &opts, sizeof(opts))) &&
-					(opts.SamplerFeedbackTier >= D3D12_SAMPLER_FEEDBACK_TIER_0_9) &&
-					check_vulkan_supported())
+				return GSRendererType::DX12;
+				/*
+				// Keep the old code as a reference if we need it.
+				const auto device12 = get_d3d12_device();
+				if (device12)
 				{
-					Console.WriteLn("Sampler feedback tier 0.9 found for Intel GPU, defaulting to Vulkan.");
-					return GSRendererType::VK;
+					D3D12_FEATURE_DATA_D3D12_OPTIONS7 opts = {};
+					if (SUCCEEDED(device12->CheckFeatureSupport(D3D12_FEATURE_D3D12_OPTIONS7, &opts, sizeof(opts))) &&
+						(opts.SamplerFeedbackTier >= D3D12_SAMPLER_FEEDBACK_TIER_0_9) &&
+						check_vulkan_supported())
+					{
+						Console.WriteLn("Sampler feedback tier 0.9 found for Intel GPU, defaulting to Vulkan.");
+						return GSRendererType::VK;
+					}
+					else
+						return GSRendererType::OGL;
 				}
+				
 				else
-				{
-					Console.WriteLn("Sampler feedback tier 0.9 or Vulkan not found for Intel GPU, using OpenGL.");
 					return GSRendererType::OGL;
-				}
+				*/
 			}
-
-			Console.WriteLn("Sampler feedback tier 0.9 or Direct3D 12 not found for Intel GPU, using Direct3D 11.");
-			return GSRendererType::DX11;
+			else if (feature_level == D3D_FEATURE_LEVEL_11_1)
+				return GSRendererType::OGL;
+			else
+				return GSRendererType::DX11;
 		}
 		break;
 

pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.cpp

@@ -724,9 +724,9 @@ bool GSDeviceOGL::CreateTextureFX()
 
 bool GSDeviceOGL::CheckFeatures()
 {
-	//bool vendor_id_amd = false;
-	//bool vendor_id_nvidia = false;
-	//bool vendor_id_intel = false;
+	bool vendor_id_amd = false;
+	bool vendor_id_nvidia = false;
+	bool vendor_id_intel = false;
 
 	memset(&m_bugs, 0, sizeof(m_bugs));
 
@@ -735,18 +735,18 @@ bool GSDeviceOGL::CheckFeatures()
 		std::strstr(vendor, "ATI"))
 	{
 		Console.WriteLn(Color_StrongRed, "GL: AMD GPU detected.");
-		//vendor_id_amd = true;
+		vendor_id_amd = true;
 	}
 	else if (std::strstr(vendor, "NVIDIA Corporation"))
 	{
 		Console.WriteLn(Color_StrongGreen, "GL: NVIDIA GPU detected.");
-		//vendor_id_nvidia = true;
+		vendor_id_nvidia = true;
 		m_bugs.broken_blend_coherency = true;
 	}
 	else if (std::strstr(vendor, "Intel"))
 	{
 		Console.WriteLn(Color_StrongBlue, "GL: Intel GPU detected.");
-		//vendor_id_intel = true;
+		vendor_id_intel = true;
 	}
 
 	GLint major_gl = 0;
@@ -906,15 +906,27 @@ bool GSDeviceOGL::CheckFeatures()
 	glGetIntegerv(GL_MAX_TEXTURE_SIZE, &max_texture_size);
 	m_max_texture_size = std::max(1024u, static_cast<u32>(max_texture_size));
 
-	Console.WriteLn("GL: Using %s for point expansion, %s for line expansion and %s for sprite expansion.",
-		m_features.point_expand ? "hardware" : (m_features.vs_expand ? "vertex expanding" : "UNSUPPORTED"),
-		m_features.line_expand ? "hardware" : (m_features.vs_expand ? "vertex expanding" : "UNSUPPORTED"),
-		m_features.vs_expand ? "vertex expanding" : "CPU");
+	// Unlikely to be supported on Windows if device is stuck on GL 3.3 which is usually equivalent to feature level 10.0.
+	// This should also target any proprietary drivers on linux.
+	// Open source drivers shouldn't be hit since they have different vendor names.
+	if ((vendor_id_amd || vendor_id_nvidia || vendor_id_intel) && GLAD_GL_VERSION_3_3 && !GLAD_GL_VERSION_4_0)
+		m_rgba16_unorm_hw_blend = false;
+	else
+		m_rgba16_unorm_hw_blend = true;
 
 	if (!GLAD_GL_ARB_conservative_depth)
 	{
 		Console.Warning("GLAD_GL_ARB_conservative_depth is not supported. This will reduce performance.");
 	}
+
+	Console.WriteLn("GL: Using %s for point expansion, %s for line expansion and %s for sprite expansion.",
+		m_features.point_expand ? "hardware" : (m_features.vs_expand ? "vertex expanding" : "UNSUPPORTED"),
+		m_features.line_expand ? "hardware" : (m_features.vs_expand ? "vertex expanding" : "UNSUPPORTED"),
+		m_features.vs_expand ? "vertex expanding" : "CPU");
+	
+	Console.WriteLnFmt("GL: DXTn Texture Compression: {}", m_features.dxt_textures ? "Supported" : "Not Supported");
+	Console.WriteLnFmt("GL: BC6/7 Texture Compression: {}", m_features.bptc_textures ? "Supported" : "Not Supported");
+	Console.WriteLnFmt("GL: RGBA16 UNORM Hardware Blending: {}", m_rgba16_unorm_hw_blend ? "Supported" : "Not Supported");
 	
 	m_features.aa1 = GSConfig.HWAA1 && m_features.vs_expand && m_features.feedback_loops();
 	
@@ -2898,7 +2910,7 @@ void GSDeviceOGL::RenderHW(GSHWDrawConfig& config)
 		{
 			config.colclip_update_area = config.drawarea;
 
-			colclip_rt = CreateFeedbackTarget(rtsize.x, rtsize.y, GSTexture::Format::ColorClip, false);
+			colclip_rt = CreateFeedbackTarget(rtsize.x, rtsize.y, m_rgba16_unorm_hw_blend ? GSTexture::Format::ColorClip : GSTexture::Format::ColorHDR, false);
 
 			if (!colclip_rt)
 			{

pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.h

@@ -159,6 +159,7 @@ private:
 	} m_bugs;
 
 	bool m_disable_download_pbo = false;
+	bool m_rgba16_unorm_hw_blend = false;
 
 	GLuint m_fbo = 0; // frame buffer container
 	GLuint m_fbo_read = 0; // frame buffer container only for reading

pcsx2/GS/Renderers/OpenGL/GSTextureOGL.cpp

@@ -34,7 +34,7 @@ GSTextureOGL::GSTextureOGL(Usage usage, int width, int height, int levels, Forma
 	// Bunch of constant parameter
 	switch (m_format)
 	{
-		// 1 Channel integer
+		// 1 channel integer
 		case Format::PrimID:
 			m_gl_format = GL_R32F;
 			m_int_format = GL_RED;
@@ -54,7 +54,7 @@ GSTextureOGL::GSTextureOGL(Usage usage, int width, int height, int levels, Forma
 			m_int_shift = 1;
 			break;
 
-		// 1 Channel normalized
+		// 1 channel normalized
 		case Format::UNorm8:
 			m_gl_format = GL_R8;
 			m_int_format = GL_RED;
@@ -72,15 +72,29 @@ GSTextureOGL::GSTextureOGL(Usage usage, int width, int height, int levels, Forma
 
 		// 4 channel normalized
 		case Format::Color:
-		case Format::ColorHQ:
-		case Format::ColorHDR:
 			m_gl_format = GL_RGBA8;
 			m_int_format = GL_RGBA;
 			m_int_type = GL_UNSIGNED_BYTE;
 			m_int_shift = 2;
 			break;
 
+		// 4 channel normalized with 2 bits of alpha
+		case Format::ColorHQ:
+			m_gl_format = GL_RGB10_A2;
+			m_int_format = GL_RGBA;
+			m_int_type = GL_UNSIGNED_INT_2_10_10_10_REV;
+			m_int_shift = 2;
+			break;
+
 		// 4 channel float
+		case Format::ColorHDR:
+			m_gl_format = GL_RGBA16F;
+			m_int_format = GL_RGBA;
+			m_int_type = GL_HALF_FLOAT;
+			m_int_shift = 3;
+			break;
+
+		// 4 channel normalized  
 		case Format::ColorClip:
 			m_gl_format = GL_RGBA16;
 			m_int_format = GL_RGBA;

pcsx2/Vif_Unpack.cpp

@@ -361,7 +361,7 @@ _vifT int nVifUnpack(const u8* data)
 				_nVifUnpack(idx, data, vifRegs.mode, isFill);
 		}
 		else
-			vu1Thread.VifUnpack(vif, vifRegs, (u8*)data, (size + 4) & ~0x3);
+			vu1Thread.VifUnpack(vif, vifRegs, (u8*)data, size);
 
 		vif.pass     = 0;
 		vif.tag.size = 0;