GS: Add option to disable vertex shader expand

And automatically disable it on Fermi (buggy driver).

pcsx2-qt/Settings/GraphicsSettingsWidget.cpp

@@ -252,6 +252,7 @@ GraphicsSettingsWidget::GraphicsSettingsWidget(SettingsDialog* dialog, QWidget*
 	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_ui.disableFramebufferFetch, "EmuCore/GS", "DisableFramebufferFetch", false);
 	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_ui.disableDualSource, "EmuCore/GS", "DisableDualSourceBlend", false);
 	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_ui.disableShaderCache, "EmuCore/GS", "DisableShaderCache", false);
+	SettingWidgetBinder::BindWidgetToBoolSetting(sif, m_ui.disableVertexShaderExpand, "EmuCore/GS", "DisableVertexShaderExpand", false);
 	SettingWidgetBinder::BindWidgetToIntSetting(
 		sif, m_ui.gsDownloadMode, "EmuCore/GS", "HWDownloadMode", static_cast<int>(GSHardwareDownloadMode::Enabled));
 

pcsx2-qt/Settings/GraphicsSettingsWidget.ui

@@ -7,7 +7,7 @@
     <x>0</x>
     <y>0</y>
     <width>720</width>
-    <height>498</height>
+    <height>492</height>
    </rect>
   </property>
   <property name="windowTitle">
@@ -2146,13 +2146,6 @@
           </item>
           <item row="2" column="0" colspan="2">
            <layout class="QGridLayout" name="gridLayout_7">
-            <item row="0" column="0">
-             <widget class="QCheckBox" name="disableDualSource">
-              <property name="text">
-               <string>Disable Dual Source Blending</string>
-              </property>
-             </widget>
-            </item>
             <item row="0" column="1">
              <widget class="QCheckBox" name="disableFramebufferFetch">
               <property name="text">
@@ -2160,10 +2153,10 @@
               </property>
              </widget>
             </item>
-            <item row="1" column="1">
-             <widget class="QCheckBox" name="useDebugDevice">
+            <item row="0" column="0">
+             <widget class="QCheckBox" name="disableDualSource">
               <property name="text">
-               <string>Use Debug Device</string>
+               <string>Disable Dual Source Blending</string>
               </property>
              </widget>
             </item>
@@ -2174,6 +2167,20 @@
               </property>
              </widget>
             </item>
+            <item row="2" column="0">
+             <widget class="QCheckBox" name="useDebugDevice">
+              <property name="text">
+               <string>Use Debug Device</string>
+              </property>
+             </widget>
+            </item>
+            <item row="1" column="1">
+             <widget class="QCheckBox" name="disableVertexShaderExpand">
+              <property name="text">
+               <string>Disable Vertex Shader Expand</string>
+              </property>
+             </widget>
+            </item>
            </layout>
           </item>
          </layout>

pcsx2/Config.h

@@ -647,6 +647,7 @@ struct Pcsx2Config
 					DisableShaderCache : 1,
 					DisableDualSourceBlend : 1,
 					DisableFramebufferFetch : 1,
+					DisableVertexShaderExpand : 1,
 					DisableThreadedPresentation : 1,
 					SkipDuplicateFrames : 1,
 					OsdShowMessages : 1,

pcsx2/GS/Renderers/DX11/GSDevice11.cpp

@@ -188,7 +188,7 @@ bool GSDevice11::Create()
 		}
 	}
 
-	SetFeatures();
+	SetFeatures(dxgi_adapter.get());
 
 	std::optional<std::string> shader = Host::ReadResourceFileToString("shaders/dx11/tfx.fx");
 	if (!shader.has_value())
@@ -532,7 +532,7 @@ void GSDevice11::Destroy()
 #endif
 }
 
-void GSDevice11::SetFeatures()
+void GSDevice11::SetFeatures(IDXGIAdapter1* adapter)
 {
 	// Check all three formats, since the feature means any can be used.
 	m_features.dxt_textures = SupportsTextureFormat(m_dev.get(), DXGI_FORMAT_BC1_UNORM) &&
@@ -542,7 +542,21 @@ void GSDevice11::SetFeatures()
 	m_features.bptc_textures = SupportsTextureFormat(m_dev.get(), DXGI_FORMAT_BC7_UNORM);
 
 	const D3D_FEATURE_LEVEL feature_level = m_dev->GetFeatureLevel();
-	m_features.vs_expand = (feature_level >= D3D_FEATURE_LEVEL_11_0);
+	m_features.vs_expand = (!GSConfig.DisableVertexShaderExpand && feature_level >= D3D_FEATURE_LEVEL_11_0);
+
+	// NVIDIA GPUs prior to Kepler appear to have broken vertex shader buffer loading.
+	if (m_features.vs_expand && (D3D::GetVendorID(adapter) == D3D::VendorID::Nvidia))
+	{
+		// There's nothing Fermi specific which we can query in DX11. Closest we have is typed UAV loads,
+		// which is Kepler+. Anyone using Kepler should be using Vulkan anyway.
+		D3D11_FEATURE_DATA_D3D11_OPTIONS2 options;
+		if (SUCCEEDED(m_dev->CheckFeatureSupport(D3D11_FEATURE_D3D11_OPTIONS2, &options, sizeof(options))) &&
+			!options.TypedUAVLoadAdditionalFormats)
+		{
+			Console.Warning("Disabling VS expand due to potentially buggy NVIDIA driver.");
+			m_features.vs_expand = false;
+		}
+	}
 }
 
 bool GSDevice11::HasSurface() const

pcsx2/GS/Renderers/DX11/GSDevice11.h

@@ -118,7 +118,7 @@ private:
 		NUM_TIMESTAMP_QUERIES = 5,
 	};
 
-	void SetFeatures();
+	void SetFeatures(IDXGIAdapter1* adapter);
 
 	bool CreateSwapChain();
 	bool CreateSwapChainRTV();

pcsx2/GS/Renderers/DX12/GSDevice12.cpp

@@ -610,7 +610,7 @@ bool GSDevice12::CheckFeatures()
 	m_features.clip_control = true;
 	m_features.stencil_buffer = true;
 	m_features.test_and_sample_depth = false;
-	m_features.vs_expand = true;
+	m_features.vs_expand = !GSConfig.DisableVertexShaderExpand;
 
 	m_features.dxt_textures = g_d3d12_context->SupportsTextureFormat(DXGI_FORMAT_BC1_UNORM) &&
 							  g_d3d12_context->SupportsTextureFormat(DXGI_FORMAT_BC2_UNORM) &&

pcsx2/GS/Renderers/Metal/GSDeviceMTL.mm

@@ -833,7 +833,7 @@ bool GSDeviceMTL::Create()
 	MTLPixelFormat layer_px_fmt = [m_layer pixelFormat];
 
 	m_features.broken_point_sampler = [[m_dev.dev name] containsString:@"AMD"];
-	m_features.vs_expand = true;
+	m_features.vs_expand = !GSConfig.DisableVertexShaderExpand;
 	m_features.primitive_id = m_dev.features.primid;
 	m_features.texture_barrier = true;
 	m_features.provoking_vertex_last = false;

pcsx2/GS/Renderers/OpenGL/GSDeviceOGL.cpp

@@ -159,12 +159,20 @@ bool GSDeviceOGL::Create()
 	m_features.stencil_buffer = true;
 	m_features.test_and_sample_depth = m_features.texture_barrier;
 
+	// NVIDIA GPUs prior to Kepler appear to have broken vertex shader buffer loading.
+	// Use bindless textures (introduced in Kepler) to differentiate.
+	const bool buggy_vs_expand =
+		GLLoader::vendor_id_nvidia && (!GLAD_GL_ARB_bindless_texture && !GLAD_GL_NV_bindless_texture);
+	if (buggy_vs_expand)
+		Console.Warning("Disabling vertex shader expand due to broken NVIDIA driver.");
+
 	if (GLAD_GL_ARB_shader_storage_buffer_object)
 	{
 		GLint max_vertex_ssbos = 0;
 		glGetIntegerv(GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS, &max_vertex_ssbos);
 		DevCon.WriteLn("GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS: %d", max_vertex_ssbos);
-		m_features.vs_expand = (max_vertex_ssbos > 0 && GLAD_GL_ARB_gpu_shader5);
+		m_features.vs_expand = (!GSConfig.DisableVertexShaderExpand && !buggy_vs_expand && max_vertex_ssbos > 0 &&
+								GLAD_GL_ARB_gpu_shader5);
 	}
 	if (!m_features.vs_expand)
 		Console.Warning("Vertex expansion is not supported. This will reduce performance.");

pcsx2/GS/Renderers/Vulkan/GSDeviceVK.cpp

@@ -696,7 +696,8 @@ bool GSDeviceVK::CheckFeatures()
 	m_features.provoking_vertex_last = g_vulkan_context->GetOptionalExtensions().vk_ext_provoking_vertex;
 	m_features.dual_source_blend = features.dualSrcBlend && !GSConfig.DisableDualSourceBlend;
 	m_features.clip_control = true;
-	m_features.vs_expand = g_vulkan_context->GetOptionalExtensions().vk_khr_shader_draw_parameters;
+	m_features.vs_expand =
+		!GSConfig.DisableVertexShaderExpand && g_vulkan_context->GetOptionalExtensions().vk_khr_shader_draw_parameters;
 
 	if (!m_features.dual_source_blend)
 		Console.Warning("Vulkan driver is missing dual-source blending. This will have an impact on performance.");

pcsx2/GS/Renderers/Vulkan/VKContext.h

@@ -52,7 +52,6 @@ public:
 		bool vk_ext_line_rasterization : 1;
 		bool vk_ext_rasterization_order_attachment_access : 1;
 		bool vk_ext_full_screen_exclusive : 1;
-		bool vk_ext_surface_maintenance1 : 1;
 		bool vk_khr_driver_properties : 1;
 		bool vk_khr_fragment_shader_barycentric : 1;
 		bool vk_khr_shader_draw_parameters : 1;

pcsx2/Pcsx2Config.cpp

@@ -415,6 +415,7 @@ Pcsx2Config::GSOptions::GSOptions()
 	UseBlitSwapChain = false;
 	DisableShaderCache = false;
 	DisableFramebufferFetch = false;
+	DisableVertexShaderExpand = false;
 	DisableThreadedPresentation = false;
 	SkipDuplicateFrames = false;
 	OsdShowMessages = true;
@@ -579,6 +580,7 @@ bool Pcsx2Config::GSOptions::RestartOptionsAreEqual(const GSOptions& right) cons
 		   OpEqu(DisableShaderCache) &&
 		   OpEqu(DisableDualSourceBlend) &&
 		   OpEqu(DisableFramebufferFetch) &&
+		   OpEqu(DisableVertexShaderExpand) &&
 		   OpEqu(DisableThreadedPresentation) &&
 		   OpEqu(OverrideTextureBarriers) &&
 		   OpEqu(ExclusiveFullscreenControl);
@@ -634,6 +636,7 @@ void Pcsx2Config::GSOptions::LoadSave(SettingsWrapper& wrap)
 	GSSettingBool(DisableShaderCache);
 	GSSettingBool(DisableDualSourceBlend);
 	GSSettingBool(DisableFramebufferFetch);
+	GSSettingBool(DisableVertexShaderExpand);
 	GSSettingBool(DisableThreadedPresentation);
 	GSSettingBool(SkipDuplicateFrames);
 	GSSettingBool(OsdShowMessages);