vk_device: Misc changes

- Setup more features and requirements.
- Improve logging for missing features.
- Collect telemetry parameters.
- Add queries for more image formats.
- Query push constants limits.
- Optionally enable some extensions.

.ci/scripts/windows/upload.ps1

@@ -39,6 +39,7 @@ mkdir "artifacts"
 # Build a tar.xz for the source of the release
 Copy-Item .\license.txt -Destination $MSVC_SOURCE
 Copy-Item .\README.md -Destination $MSVC_SOURCE
+Copy-Item .\CMakeLists.txt -Destination $MSVC_SOURCE
 Copy-Item .\src -Recurse -Destination $MSVC_SOURCE
 Copy-Item .\externals -Recurse -Destination $MSVC_SOURCE
 Copy-Item .\dist -Recurse -Destination $MSVC_SOURCE
@@ -60,4 +61,4 @@ Get-ChildItem "$BUILD_DIR" -Recurse -Filter "QtWebEngineProcess*.exe" | Copy-Ite
 
 Get-ChildItem . -Filter "*.zip" | Copy-Item -destination "artifacts"
 Get-ChildItem . -Filter "*.7z" | Copy-Item -destination "artifacts"
-Get-ChildItem . -Filter "*.tar.xz" | Copy-Item -destination "artifacts"
+Get-ChildItem . -Filter "*.tar.xz" | Copy-Item -destination "artifacts"

CMakeLists.txt

@@ -29,7 +29,7 @@ option(ENABLE_VULKAN "Enables Vulkan backend" ON)
 
 option(USE_DISCORD_PRESENCE "Enables Discord Rich Presence" OFF)
 
-if(NOT EXISTS ${PROJECT_SOURCE_DIR}/.git/hooks/pre-commit)
+if(EXISTS ${PROJECT_SOURCE_DIR}/hooks/pre-commit AND NOT EXISTS ${PROJECT_SOURCE_DIR}/.git/hooks/pre-commit)
     message(STATUS "Copying pre-commit hook")
     file(COPY hooks/pre-commit
         DESTINATION ${PROJECT_SOURCE_DIR}/.git/hooks)
@@ -49,7 +49,10 @@ function(check_submodules_present)
         endif()
     endforeach()
 endfunction()
-check_submodules_present()
+
+if(EXISTS ${PROJECT_SOURCE_DIR}/.gitmodules)
+    check_submodules_present()
+endif()
 
 configure_file(${PROJECT_SOURCE_DIR}/dist/compatibility_list/compatibility_list.qrc
                ${PROJECT_BINARY_DIR}/dist/compatibility_list/compatibility_list.qrc

src/core/core_cpu.cpp

@@ -96,6 +96,8 @@ void Cpu::RunLoop(bool tight_loop) {
         } else {
             arm_interface->Step();
         }
+        // We are stopping a run, exclusive state must be cleared
+        arm_interface->ClearExclusiveState();
     }
     core_timing.Advance();
 

src/core/hle/kernel/address_arbiter.cpp

@@ -11,8 +11,6 @@
 #include "core/core_cpu.h"
 #include "core/hle/kernel/address_arbiter.h"
 #include "core/hle/kernel/errors.h"
-#include "core/hle/kernel/object.h"
-#include "core/hle/kernel/process.h"
 #include "core/hle/kernel/scheduler.h"
 #include "core/hle/kernel/thread.h"
 #include "core/hle/result.h"

src/core/hle/kernel/address_arbiter.h

@@ -4,10 +4,10 @@
 
 #pragma once
 
+#include <memory>
 #include <vector>
 
 #include "common/common_types.h"
-#include "core/hle/kernel/object.h"
 
 union ResultCode;
 

src/core/hle/kernel/client_port.cpp

@@ -8,7 +8,6 @@
 #include "core/hle/kernel/hle_ipc.h"
 #include "core/hle/kernel/object.h"
 #include "core/hle/kernel/server_port.h"
-#include "core/hle/kernel/server_session.h"
 #include "core/hle/kernel/session.h"
 
 namespace Kernel {

src/core/hle/kernel/client_port.h

@@ -4,7 +4,9 @@
 
 #pragma once
 
+#include <memory>
 #include <string>
+
 #include "common/common_types.h"
 #include "core/hle/kernel/object.h"
 #include "core/hle/result.h"

src/core/hle/kernel/handle_table.h

@@ -6,6 +6,8 @@
 
 #include <array>
 #include <cstddef>
+#include <memory>
+
 #include "common/common_types.h"
 #include "core/hle/kernel/object.h"
 #include "core/hle/result.h"

src/core/hle/kernel/kernel.cpp

@@ -13,7 +13,6 @@
 #include "core/core.h"
 #include "core/core_timing.h"
 #include "core/core_timing_util.h"
-#include "core/hle/kernel/address_arbiter.h"
 #include "core/hle/kernel/client_port.h"
 #include "core/hle/kernel/errors.h"
 #include "core/hle/kernel/handle_table.h"

src/core/hle/kernel/kernel.h

@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include <memory>
 #include <string>
 #include <unordered_map>
 #include <vector>

src/core/hle/kernel/mutex.cpp

@@ -2,6 +2,7 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <memory>
 #include <utility>
 #include <vector>
 

src/core/hle/kernel/resource_limit.h

@@ -5,6 +5,8 @@
 #pragma once
 
 #include <array>
+#include <memory>
+
 #include "common/common_types.h"
 #include "core/hle/kernel/object.h"
 

src/core/hle/kernel/scheduler.cpp

@@ -458,7 +458,6 @@ void Scheduler::SwitchContext() {
         cpu_core.LoadContext(new_thread->GetContext());
         cpu_core.SetTlsAddress(new_thread->GetTLSAddress());
         cpu_core.SetTPIDR_EL0(new_thread->GetTPIDR_EL0());
-        cpu_core.ClearExclusiveState();
     } else {
         current_thread = nullptr;
         // Note: We do not reset the current process and current page table when idling because

src/core/hle/kernel/scheduler.h

@@ -4,11 +4,12 @@
 
 #pragma once
 
-#include <mutex>
+#include <atomic>
+#include <memory>
 #include <vector>
+
 #include "common/common_types.h"
 #include "common/multi_level_queue.h"
-#include "core/hle/kernel/object.h"
 #include "core/hle/kernel/thread.h"
 
 namespace Core {

src/core/hle/kernel/session.h

@@ -6,9 +6,9 @@
 
 #include <memory>
 #include <string>
+#include <utility>
 
 #include "core/hle/kernel/wait_object.h"
-#include "core/hle/result.h"
 
 namespace Kernel {
 

src/core/hle/kernel/shared_memory.h

@@ -6,7 +6,6 @@
 
 #include <memory>
 #include <string>
-#include <vector>
 
 #include "common/common_types.h"
 #include "core/hle/kernel/object.h"

src/core/hle/kernel/transfer_memory.h

@@ -5,7 +5,6 @@
 #pragma once
 
 #include <memory>
-#include <vector>
 
 #include "core/hle/kernel/object.h"
 #include "core/hle/kernel/physical_memory.h"

src/core/hle/kernel/wait_object.h

@@ -4,8 +4,9 @@
 
 #pragma once
 
+#include <memory>
 #include <vector>
-#include <boost/smart_ptr/intrusive_ptr.hpp>
+
 #include "core/hle/kernel/object.h"
 
 namespace Kernel {

src/core/hle/kernel/writable_event.h

@@ -4,6 +4,8 @@
 
 #pragma once
 
+#include <memory>
+
 #include "core/hle/kernel/object.h"
 
 namespace Kernel {

src/core/telemetry_session.cpp

@@ -165,24 +165,20 @@ void TelemetrySession::AddInitialInfo(Loader::AppLoader& app_loader) {
     Telemetry::AppendOSInfo(field_collection);
 
     // Log user configuration information
-    AddField(Telemetry::FieldType::UserConfig, "Audio_SinkId", Settings::values.sink_id);
-    AddField(Telemetry::FieldType::UserConfig, "Audio_EnableAudioStretching",
-             Settings::values.enable_audio_stretching);
-    AddField(Telemetry::FieldType::UserConfig, "Core_UseMultiCore",
-             Settings::values.use_multi_core);
-    AddField(Telemetry::FieldType::UserConfig, "Renderer_ResolutionFactor",
-             Settings::values.resolution_factor);
-    AddField(Telemetry::FieldType::UserConfig, "Renderer_UseFrameLimit",
-             Settings::values.use_frame_limit);
-    AddField(Telemetry::FieldType::UserConfig, "Renderer_FrameLimit", Settings::values.frame_limit);
-    AddField(Telemetry::FieldType::UserConfig, "Renderer_UseDiskShaderCache",
-             Settings::values.use_disk_shader_cache);
-    AddField(Telemetry::FieldType::UserConfig, "Renderer_UseAccurateGpuEmulation",
+    constexpr auto field_type = Telemetry::FieldType::UserConfig;
+    AddField(field_type, "Audio_SinkId", Settings::values.sink_id);
+    AddField(field_type, "Audio_EnableAudioStretching", Settings::values.enable_audio_stretching);
+    AddField(field_type, "Core_UseMultiCore", Settings::values.use_multi_core);
+    AddField(field_type, "Renderer_Backend", "OpenGL");
+    AddField(field_type, "Renderer_ResolutionFactor", Settings::values.resolution_factor);
+    AddField(field_type, "Renderer_UseFrameLimit", Settings::values.use_frame_limit);
+    AddField(field_type, "Renderer_FrameLimit", Settings::values.frame_limit);
+    AddField(field_type, "Renderer_UseDiskShaderCache", Settings::values.use_disk_shader_cache);
+    AddField(field_type, "Renderer_UseAccurateGpuEmulation",
              Settings::values.use_accurate_gpu_emulation);
-    AddField(Telemetry::FieldType::UserConfig, "Renderer_UseAsynchronousGpuEmulation",
+    AddField(field_type, "Renderer_UseAsynchronousGpuEmulation",
              Settings::values.use_asynchronous_gpu_emulation);
-    AddField(Telemetry::FieldType::UserConfig, "System_UseDockedMode",
-             Settings::values.use_docked_mode);
+    AddField(field_type, "System_UseDockedMode", Settings::values.use_docked_mode);
 }
 
 bool TelemetrySession::SubmitTestcase() {

src/video_core/engines/maxwell_3d.h

@@ -491,6 +491,23 @@ public:
             INSERT_UNION_PADDING_WORDS(1);
         };
 
+        enum class DepthMode : u32 {
+            MinusOneToOne = 0,
+            ZeroToOne = 1,
+        };
+
+        enum class TessellationPrimitive : u32 {
+            Isolines = 0,
+            Triangles = 1,
+            Quads = 2,
+        };
+
+        enum class TessellationSpacing : u32 {
+            Equal = 0,
+            FractionalOdd = 1,
+            FractionalEven = 2,
+        };
+
         struct RenderTargetConfig {
             u32 address_high;
             u32 address_low;
@@ -628,7 +645,19 @@ public:
                     };
                 } sync_info;
 
-                INSERT_UNION_PADDING_WORDS(0x11E);
+                INSERT_UNION_PADDING_WORDS(0x15);
+
+                union {
+                    BitField<0, 2, TessellationPrimitive> prim;
+                    BitField<4, 2, TessellationSpacing> spacing;
+                    BitField<8, 1, u32> cw;
+                    BitField<9, 1, u32> connected;
+                } tess_mode;
+
+                std::array<f32, 4> tess_level_outer;
+                std::array<f32, 2> tess_level_inner;
+
+                INSERT_UNION_PADDING_WORDS(0x102);
 
                 u32 tfb_enabled;
 
@@ -662,7 +691,9 @@ public:
                 u32 polygon_offset_line_enable;
                 u32 polygon_offset_fill_enable;
 
-                INSERT_UNION_PADDING_WORDS(0xD);
+                u32 patch_vertices;
+
+                INSERT_UNION_PADDING_WORDS(0xC);
 
                 std::array<ScissorTest, NumViewports> scissor_test;
 
@@ -1386,6 +1417,9 @@ ASSERT_REG_POSITION(upload, 0x60);
 ASSERT_REG_POSITION(exec_upload, 0x6C);
 ASSERT_REG_POSITION(data_upload, 0x6D);
 ASSERT_REG_POSITION(sync_info, 0xB2);
+ASSERT_REG_POSITION(tess_mode, 0xC8);
+ASSERT_REG_POSITION(tess_level_outer, 0xC9);
+ASSERT_REG_POSITION(tess_level_inner, 0xCD);
 ASSERT_REG_POSITION(tfb_enabled, 0x1D1);
 ASSERT_REG_POSITION(rt, 0x200);
 ASSERT_REG_POSITION(viewport_transform, 0x280);
@@ -1397,6 +1431,7 @@ ASSERT_REG_POSITION(clear_stencil, 0x368);
 ASSERT_REG_POSITION(polygon_offset_point_enable, 0x370);
 ASSERT_REG_POSITION(polygon_offset_line_enable, 0x371);
 ASSERT_REG_POSITION(polygon_offset_fill_enable, 0x372);
+ASSERT_REG_POSITION(patch_vertices, 0x373);
 ASSERT_REG_POSITION(scissor_test, 0x380);
 ASSERT_REG_POSITION(stencil_back_func_ref, 0x3D5);
 ASSERT_REG_POSITION(stencil_back_func_mask, 0x3D6);

src/video_core/engines/shader_bytecode.h

@@ -799,6 +799,12 @@ union Instruction {
         BitField<40, 1, u64> invert;
     } popc;
 
+    union {
+        BitField<41, 1, u64> sh;
+        BitField<40, 1, u64> invert;
+        BitField<48, 1, u64> is_signed;
+    } flo;
+
     union {
         BitField<39, 3, u64> pred;
         BitField<42, 1, u64> neg_pred;
@@ -1439,6 +1445,26 @@ union Instruction {
         }
     } tlds;
 
+    union {
+        BitField<28, 1, u64> is_array;
+        BitField<29, 2, TextureType> texture_type;
+        BitField<35, 1, u64> aoffi_flag;
+        BitField<49, 1, u64> nodep_flag;
+
+        bool UsesMiscMode(TextureMiscMode mode) const {
+            switch (mode) {
+            case TextureMiscMode::AOFFI:
+                return aoffi_flag != 0;
+            case TextureMiscMode::NODEP:
+                return nodep_flag != 0;
+            default:
+                break;
+            }
+            return false;
+        }
+
+    } txd;
+
     union {
         BitField<24, 2, StoreCacheManagement> cache_management;
         BitField<33, 3, ImageType> image_type;
@@ -1632,6 +1658,8 @@ public:
         TLD4S,  // Texture Load 4 with scalar / non - vec4 source / destinations
         TMML_B, // Texture Mip Map Level
         TMML,   // Texture Mip Map Level
+        TXD,    // Texture Gradient/Load with Derivates
+        TXD_B,  // Texture Gradient/Load with Derivates Bindless
         SUST,   // Surface Store
         SULD,   // Surface Load
         SUATOM, // Surface Atomic Operation
@@ -1664,6 +1692,9 @@ public:
         ISCADD_C, // Scale and Add
         ISCADD_R,
         ISCADD_IMM,
+        FLO_R,
+        FLO_C,
+        FLO_IMM,
         LEA_R1,
         LEA_R2,
         LEA_RZ,
@@ -1727,6 +1758,10 @@ public:
         SHR_C,
         SHR_R,
         SHR_IMM,
+        SHF_RIGHT_R,
+        SHF_RIGHT_IMM,
+        SHF_LEFT_R,
+        SHF_LEFT_IMM,
         FMNMX_C,
         FMNMX_R,
         FMNMX_IMM,
@@ -1924,6 +1959,8 @@ private:
             INST("1101111100------", Id::TLD4S, Type::Texture, "TLD4S"),
             INST("110111110110----", Id::TMML_B, Type::Texture, "TMML_B"),
             INST("1101111101011---", Id::TMML, Type::Texture, "TMML"),
+            INST("11011110011110--", Id::TXD_B, Type::Texture, "TXD_B"),
+            INST("11011110001110--", Id::TXD, Type::Texture, "TXD"),
             INST("11101011001-----", Id::SUST, Type::Image, "SUST"),
             INST("11101011000-----", Id::SULD, Type::Image, "SULD"),
             INST("1110101000------", Id::SUATOM, Type::Image, "SUATOM_D"),
@@ -1965,6 +2002,9 @@ private:
             INST("010110110100----", Id::ICMP_R, Type::ArithmeticInteger, "ICMP_R"),
             INST("010010110100----", Id::ICMP_CR, Type::ArithmeticInteger, "ICMP_CR"),
             INST("0011011-0100----", Id::ICMP_IMM, Type::ArithmeticInteger, "ICMP_IMM"),
+            INST("0101110000110---", Id::FLO_R, Type::ArithmeticInteger, "FLO_R"),
+            INST("0100110000110---", Id::FLO_C, Type::ArithmeticInteger, "FLO_C"),
+            INST("0011100-00110---", Id::FLO_IMM, Type::ArithmeticInteger, "FLO_IMM"),
             INST("0101101111011---", Id::LEA_R2, Type::ArithmeticInteger, "LEA_R2"),
             INST("0101101111010---", Id::LEA_R1, Type::ArithmeticInteger, "LEA_R1"),
             INST("001101101101----", Id::LEA_IMM, Type::ArithmeticInteger, "LEA_IMM"),
@@ -2022,6 +2062,10 @@ private:
             INST("0100110000101---", Id::SHR_C, Type::Shift, "SHR_C"),
             INST("0101110000101---", Id::SHR_R, Type::Shift, "SHR_R"),
             INST("0011100-00101---", Id::SHR_IMM, Type::Shift, "SHR_IMM"),
+            INST("0101110011111---", Id::SHF_RIGHT_R, Type::Shift, "SHF_RIGHT_R"),
+            INST("0011100-11111---", Id::SHF_RIGHT_IMM, Type::Shift, "SHF_RIGHT_IMM"),
+            INST("0101101111111---", Id::SHF_LEFT_R, Type::Shift, "SHF_LEFT_R"),
+            INST("0011011-11111---", Id::SHF_LEFT_IMM, Type::Shift, "SHF_LEFT_IMM"),
             INST("0100110011100---", Id::I2I_C, Type::Conversion, "I2I_C"),
             INST("0101110011100---", Id::I2I_R, Type::Conversion, "I2I_R"),
             INST("0011101-11100---", Id::I2I_IMM, Type::Conversion, "I2I_IMM"),

src/video_core/renderer_opengl/gl_framebuffer_cache.cpp

@@ -3,9 +3,12 @@
 // Refer to the license.txt file included.
 
 #include <tuple>
+#include <unordered_map>
+#include <utility>
 
-#include "common/cityhash.h"
-#include "common/scope_exit.h"
+#include <glad/glad.h>
+
+#include "common/common_types.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_opengl/gl_framebuffer_cache.h"
 #include "video_core/renderer_opengl/gl_state.h"
@@ -13,6 +16,7 @@
 namespace OpenGL {
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
+using VideoCore::Surface::SurfaceType;
 
 FramebufferCacheOpenGL::FramebufferCacheOpenGL() = default;
 
@@ -35,36 +39,49 @@ OGLFramebuffer FramebufferCacheOpenGL::CreateFramebuffer(const FramebufferCacheK
     local_state.draw.draw_framebuffer = framebuffer.handle;
     local_state.ApplyFramebufferState();
 
-    for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
-        if (key.colors[index]) {
-            key.colors[index]->Attach(GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index),
-                                      GL_DRAW_FRAMEBUFFER);
-        }
-    }
-    if (key.colors_count) {
-        glDrawBuffers(key.colors_count, key.color_attachments.data());
-    } else {
-        glDrawBuffer(GL_NONE);
+    if (key.zeta) {
+        const bool stencil = key.zeta->GetSurfaceParams().type == SurfaceType::DepthStencil;
+        const GLenum attach_target = stencil ? GL_DEPTH_STENCIL_ATTACHMENT : GL_DEPTH_ATTACHMENT;
+        key.zeta->Attach(attach_target, GL_DRAW_FRAMEBUFFER);
     }
 
-    if (key.zeta) {
-        key.zeta->Attach(key.stencil_enable ? GL_DEPTH_STENCIL_ATTACHMENT : GL_DEPTH_ATTACHMENT,
-                         GL_DRAW_FRAMEBUFFER);
+    std::size_t num_buffers = 0;
+    std::array<GLenum, Maxwell::NumRenderTargets> targets;
+
+    for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
+        if (!key.colors[index]) {
+            targets[index] = GL_NONE;
+            continue;
+        }
+        const GLenum attach_target = GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index);
+        key.colors[index]->Attach(attach_target, GL_DRAW_FRAMEBUFFER);
+
+        const u32 attachment = (key.color_attachments >> (BitsPerAttachment * index)) & 0b1111;
+        targets[index] = GL_COLOR_ATTACHMENT0 + attachment;
+        num_buffers = index + 1;
+    }
+
+    if (num_buffers > 0) {
+        glDrawBuffers(static_cast<GLsizei>(num_buffers), std::data(targets));
+    } else {
+        glDrawBuffer(GL_NONE);
     }
 
     return framebuffer;
 }
 
-std::size_t FramebufferCacheKey::Hash() const {
-    static_assert(sizeof(*this) % sizeof(u64) == 0, "Unaligned struct");
-    return static_cast<std::size_t>(
-        Common::CityHash64(reinterpret_cast<const char*>(this), sizeof(*this)));
+std::size_t FramebufferCacheKey::Hash() const noexcept {
+    std::size_t hash = std::hash<View>{}(zeta);
+    for (const auto& color : colors) {
+        hash ^= std::hash<View>{}(color);
+    }
+    hash ^= static_cast<std::size_t>(color_attachments) << 16;
+    return hash;
 }
 
-bool FramebufferCacheKey::operator==(const FramebufferCacheKey& rhs) const {
-    return std::tie(stencil_enable, colors_count, color_attachments, colors, zeta) ==
-           std::tie(rhs.stencil_enable, rhs.colors_count, rhs.color_attachments, rhs.colors,
-                    rhs.zeta);
+bool FramebufferCacheKey::operator==(const FramebufferCacheKey& rhs) const noexcept {
+    return std::tie(colors, zeta, color_attachments) ==
+           std::tie(rhs.colors, rhs.zeta, rhs.color_attachments);
 }
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_framebuffer_cache.h

@@ -18,21 +18,24 @@
 
 namespace OpenGL {
 
-struct alignas(sizeof(u64)) FramebufferCacheKey {
-    bool stencil_enable = false;
-    u16 colors_count = 0;
+constexpr std::size_t BitsPerAttachment = 4;
 
-    std::array<GLenum, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> color_attachments{};
-    std::array<View, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> colors;
+struct FramebufferCacheKey {
     View zeta;
+    std::array<View, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> colors;
+    u32 color_attachments = 0;
 
-    std::size_t Hash() const;
+    std::size_t Hash() const noexcept;
 
-    bool operator==(const FramebufferCacheKey& rhs) const;
+    bool operator==(const FramebufferCacheKey& rhs) const noexcept;
 
-    bool operator!=(const FramebufferCacheKey& rhs) const {
+    bool operator!=(const FramebufferCacheKey& rhs) const noexcept {
         return !operator==(rhs);
     }
+
+    void SetAttachment(std::size_t index, u32 attachment) {
+        color_attachments |= attachment << (BitsPerAttachment * index);
+    }
 };
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -93,7 +93,6 @@ RasterizerOpenGL::RasterizerOpenGL(Core::System& system, Core::Frontend::EmuWind
     shader_program_manager = std::make_unique<GLShader::ProgramManager>();
     state.draw.shader_program = 0;
     state.Apply();
-    clear_framebuffer.Create();
 
     LOG_DEBUG(Render_OpenGL, "Sync fixed function OpenGL state here");
     CheckExtensions();
@@ -373,78 +372,58 @@ void RasterizerOpenGL::ConfigureFramebuffers() {
     UNIMPLEMENTED_IF(regs.rt_separate_frag_data == 0);
 
     // Bind the framebuffer surfaces
-    FramebufferCacheKey fbkey;
-    for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
+    FramebufferCacheKey key;
+    const auto colors_count = static_cast<std::size_t>(regs.rt_control.count);
+    for (std::size_t index = 0; index < colors_count; ++index) {
         View color_surface{texture_cache.GetColorBufferSurface(index, true)};
-
-        if (color_surface) {
-            // Assume that a surface will be written to if it is used as a framebuffer, even
-            // if the shader doesn't actually write to it.
-            texture_cache.MarkColorBufferInUse(index);
+        if (!color_surface) {
+            continue;
         }
+        // Assume that a surface will be written to if it is used as a framebuffer, even
+        // if the shader doesn't actually write to it.
+        texture_cache.MarkColorBufferInUse(index);
 
-        fbkey.color_attachments[index] = GL_COLOR_ATTACHMENT0 + regs.rt_control.GetMap(index);
-        fbkey.colors[index] = std::move(color_surface);
+        key.SetAttachment(index, regs.rt_control.GetMap(index));
+        key.colors[index] = std::move(color_surface);
     }
-    fbkey.colors_count = static_cast<u16>(regs.rt_control.count);
 
     if (depth_surface) {
         // Assume that a surface will be written to if it is used as a framebuffer, even if
         // the shader doesn't actually write to it.
         texture_cache.MarkDepthBufferInUse();
-
-        fbkey.stencil_enable = depth_surface->GetSurfaceParams().type == SurfaceType::DepthStencil;
-        fbkey.zeta = std::move(depth_surface);
+        key.zeta = std::move(depth_surface);
     }
 
     texture_cache.GuardRenderTargets(false);
 
-    state.draw.draw_framebuffer = framebuffer_cache.GetFramebuffer(fbkey);
+    state.draw.draw_framebuffer = framebuffer_cache.GetFramebuffer(key);
     SyncViewport(state);
 }
 
 void RasterizerOpenGL::ConfigureClearFramebuffer(OpenGLState& current_state, bool using_color_fb,
                                                  bool using_depth_fb, bool using_stencil_fb) {
+    using VideoCore::Surface::SurfaceType;
+
     auto& gpu = system.GPU().Maxwell3D();
     const auto& regs = gpu.regs;
 
     texture_cache.GuardRenderTargets(true);
-    View color_surface{};
+    View color_surface;
     if (using_color_fb) {
         color_surface = texture_cache.GetColorBufferSurface(regs.clear_buffers.RT, false);
     }
-    View depth_surface{};
+    View depth_surface;
     if (using_depth_fb || using_stencil_fb) {
         depth_surface = texture_cache.GetDepthBufferSurface(false);
     }
     texture_cache.GuardRenderTargets(false);
 
-    current_state.draw.draw_framebuffer = clear_framebuffer.handle;
+    FramebufferCacheKey key;
+    key.colors[0] = color_surface;
+    key.zeta = depth_surface;
+
+    current_state.draw.draw_framebuffer = framebuffer_cache.GetFramebuffer(key);
     current_state.ApplyFramebufferState();
-
-    if (color_surface) {
-        color_surface->Attach(GL_COLOR_ATTACHMENT0, GL_DRAW_FRAMEBUFFER);
-    } else {
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
-    }
-
-    if (depth_surface) {
-        const auto& params = depth_surface->GetSurfaceParams();
-        switch (params.type) {
-        case VideoCore::Surface::SurfaceType::Depth:
-            depth_surface->Attach(GL_DEPTH_ATTACHMENT, GL_DRAW_FRAMEBUFFER);
-            glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
-            break;
-        case VideoCore::Surface::SurfaceType::DepthStencil:
-            depth_surface->Attach(GL_DEPTH_STENCIL_ATTACHMENT, GL_DRAW_FRAMEBUFFER);
-            break;
-        default:
-            UNIMPLEMENTED();
-        }
-    } else {
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
-                               0);
-    }
 }
 
 void RasterizerOpenGL::Clear() {

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -223,8 +223,6 @@ private:
 
     enum class AccelDraw { Disabled, Arrays, Indexed };
     AccelDraw accelerate_draw = AccelDraw::Disabled;
-
-    OGLFramebuffer clear_framebuffer;
 };
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -49,8 +49,9 @@ class ExprDecompiler;
 enum class Type { Void, Bool, Bool2, Float, Int, Uint, HalfFloat };
 
 struct TextureAoffi {};
+struct TextureDerivates {};
 using TextureArgument = std::pair<Type, Node>;
-using TextureIR = std::variant<TextureAoffi, TextureArgument>;
+using TextureIR = std::variant<TextureAoffi, TextureDerivates, TextureArgument>;
 
 constexpr u32 MAX_CONSTBUFFER_ELEMENTS =
     static_cast<u32>(Maxwell::MaxConstBufferSize) / (4 * sizeof(float));
@@ -1112,6 +1113,8 @@ private:
                 expr += GenerateTextureArgument(*argument);
             } else if (std::holds_alternative<TextureAoffi>(variant)) {
                 expr += GenerateTextureAoffi(meta->aoffi);
+            } else if (std::holds_alternative<TextureDerivates>(variant)) {
+                expr += GenerateTextureDerivates(meta->derivates);
             } else {
                 UNREACHABLE();
             }
@@ -1181,6 +1184,36 @@ private:
         return expr;
     }
 
+    std::string GenerateTextureDerivates(const std::vector<Node>& derivates) {
+        if (derivates.empty()) {
+            return {};
+        }
+        constexpr std::array coord_constructors = {"float", "vec2", "vec3"};
+        std::string expr = ", ";
+        const std::size_t components = derivates.size() / 2;
+        std::string dx = coord_constructors.at(components - 1);
+        std::string dy = coord_constructors.at(components - 1);
+        dx += '(';
+        dy += '(';
+
+        for (std::size_t index = 0; index < components; ++index) {
+            const auto operand_x{derivates.at(index * 2)};
+            const auto operand_y{derivates.at(index * 2 + 1)};
+            dx += Visit(operand_x).AsFloat();
+            dy += Visit(operand_y).AsFloat();
+
+            if (index + 1 < components) {
+                dx += ", ";
+                dy += ", ";
+            }
+        }
+        dx += ')';
+        dy += ')';
+        expr += dx + ", " + dy;
+
+        return expr;
+    }
+
     std::string BuildIntegerCoordinates(Operation operation) {
         constexpr std::array constructors{"int(", "ivec2(", "ivec3(", "ivec4("};
         const std::size_t coords_count{operation.GetOperandsCount()};
@@ -1450,6 +1483,11 @@ private:
         return GenerateUnary(operation, "bitCount", type, type);
     }
 
+    template <Type type>
+    Expression BitMSB(Operation operation) {
+        return GenerateUnary(operation, "findMSB", type, type);
+    }
+
     Expression HNegate(Operation operation) {
         const auto GetNegate = [&](std::size_t index) {
             return VisitOperand(operation, index).AsBool() + " ? -1 : 1";
@@ -1738,6 +1776,14 @@ private:
         return {std::move(expr), Type::Float};
     }
 
+    Expression TextureGradient(Operation operation) {
+        const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
+        ASSERT(meta);
+
+        std::string expr = GenerateTexture(operation, "Grad", {TextureDerivates{}, TextureAoffi{}});
+        return {std::move(expr) + GetSwizzle(meta->element), Type::Float};
+    }
+
     Expression ImageLoad(Operation operation) {
         if (!device.HasImageLoadFormatted()) {
             LOG_ERROR(Render_OpenGL,
@@ -2003,6 +2049,7 @@ private:
         &GLSLDecompiler::BitfieldInsert<Type::Int>,
         &GLSLDecompiler::BitfieldExtract<Type::Int>,
         &GLSLDecompiler::BitCount<Type::Int>,
+        &GLSLDecompiler::BitMSB<Type::Int>,
 
         &GLSLDecompiler::Add<Type::Uint>,
         &GLSLDecompiler::Mul<Type::Uint>,
@@ -2021,6 +2068,7 @@ private:
         &GLSLDecompiler::BitfieldInsert<Type::Uint>,
         &GLSLDecompiler::BitfieldExtract<Type::Uint>,
         &GLSLDecompiler::BitCount<Type::Uint>,
+        &GLSLDecompiler::BitMSB<Type::Uint>,
 
         &GLSLDecompiler::Add<Type::HalfFloat>,
         &GLSLDecompiler::Mul<Type::HalfFloat>,
@@ -2084,6 +2132,7 @@ private:
         &GLSLDecompiler::TextureQueryDimensions,
         &GLSLDecompiler::TextureQueryLod,
         &GLSLDecompiler::TexelFetch,
+        &GLSLDecompiler::TextureGradient,
 
         &GLSLDecompiler::ImageLoad,
         &GLSLDecompiler::ImageStore,

src/video_core/renderer_vulkan/declarations.h

@@ -4,6 +4,17 @@
 
 #pragma once
 
+namespace vk {
+class DispatchLoaderDynamic;
+}
+
+namespace Vulkan {
+constexpr vk::DispatchLoaderDynamic* dont_use_me_dld = nullptr;
+}
+
+#define VULKAN_HPP_DEFAULT_DISPATCHER (*::Vulkan::dont_use_me_dld)
+#define VULKAN_HPP_ENABLE_DYNAMIC_LOADER_TOOL 0
+#define VULKAN_HPP_DISPATCH_LOADER_DYNAMIC 1
 #include <vulkan/vulkan.hpp>
 
 namespace Vulkan {
@@ -41,5 +52,7 @@ using UniqueSemaphore = UniqueHandle<vk::Semaphore>;
 using UniqueShaderModule = UniqueHandle<vk::ShaderModule>;
 using UniqueSwapchainKHR = UniqueHandle<vk::SwapchainKHR>;
 using UniqueValidationCacheEXT = UniqueHandle<vk::ValidationCacheEXT>;
+using UniqueDebugReportCallbackEXT = UniqueHandle<vk::DebugReportCallbackEXT>;
+using UniqueDebugUtilsMessengerEXT = UniqueHandle<vk::DebugUtilsMessengerEXT>;
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_device.cpp

@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.
 
 #include <bitset>
+#include <cstdlib>
 #include <optional>
 #include <set>
 #include <string_view>
@@ -15,6 +16,15 @@ namespace Vulkan {
 
 namespace {
 
+namespace Alternatives {
+
+constexpr std::array Depth24UnormS8Uint = {vk::Format::eD32SfloatS8Uint,
+                                           vk::Format::eD16UnormS8Uint, vk::Format{}};
+constexpr std::array Depth16UnormS8Uint = {vk::Format::eD24UnormS8Uint,
+                                           vk::Format::eD32SfloatS8Uint, vk::Format{}};
+
+} // namespace Alternatives
+
 template <typename T>
 void SetNext(void**& next, T& data) {
     *next = &data;
@@ -22,7 +32,7 @@ void SetNext(void**& next, T& data) {
 }
 
 template <typename T>
-T GetFeatures(vk::PhysicalDevice physical, vk::DispatchLoaderDynamic dldi) {
+T GetFeatures(vk::PhysicalDevice physical, const vk::DispatchLoaderDynamic& dldi) {
     vk::PhysicalDeviceFeatures2 features;
     T extension_features;
     features.pNext = &extension_features;
@@ -30,17 +40,14 @@ T GetFeatures(vk::PhysicalDevice physical, vk::DispatchLoaderDynamic dldi) {
     return extension_features;
 }
 
-} // Anonymous namespace
-
-namespace Alternatives {
-
-constexpr std::array Depth24UnormS8Uint = {vk::Format::eD32SfloatS8Uint,
-                                           vk::Format::eD16UnormS8Uint, vk::Format{}};
-constexpr std::array Depth16UnormS8Uint = {vk::Format::eD24UnormS8Uint,
-                                           vk::Format::eD32SfloatS8Uint, vk::Format{}};
-constexpr std::array Astc = {vk::Format::eA8B8G8R8UnormPack32, vk::Format{}};
-
-} // namespace Alternatives
+template <typename T>
+T GetProperties(vk::PhysicalDevice physical, const vk::DispatchLoaderDynamic& dldi) {
+    vk::PhysicalDeviceProperties2 properties;
+    T extension_properties;
+    properties.pNext = &extension_properties;
+    physical.getProperties2(&properties, dldi);
+    return extension_properties;
+}
 
 constexpr const vk::Format* GetFormatAlternatives(vk::Format format) {
     switch (format) {
@@ -53,8 +60,7 @@ constexpr const vk::Format* GetFormatAlternatives(vk::Format format) {
     }
 }
 
-constexpr vk::FormatFeatureFlags GetFormatFeatures(vk::FormatProperties properties,
-                                                   FormatType format_type) {
+vk::FormatFeatureFlags GetFormatFeatures(vk::FormatProperties properties, FormatType format_type) {
     switch (format_type) {
     case FormatType::Linear:
         return properties.linearTilingFeatures;
@@ -67,11 +73,13 @@ constexpr vk::FormatFeatureFlags GetFormatFeatures(vk::FormatProperties properti
     }
 }
 
+} // Anonymous namespace
+
 VKDevice::VKDevice(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical,
                    vk::SurfaceKHR surface)
-    : physical{physical}, format_properties{GetFormatProperties(dldi, physical)} {
+    : physical{physical}, properties{physical.getProperties(dldi)},
+      format_properties{GetFormatProperties(dldi, physical)} {
     SetupFamilies(dldi, surface);
-    SetupProperties(dldi);
     SetupFeatures(dldi);
 }
 
@@ -89,12 +97,22 @@ bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instan
     features.depthClamp = true;
     features.samplerAnisotropy = true;
     features.largePoints = true;
+    features.multiViewport = true;
+    features.depthBiasClamp = true;
+    features.geometryShader = true;
+    features.tessellationShader = true;
+    features.fragmentStoresAndAtomics = true;
+    features.shaderImageGatherExtended = true;
+    features.shaderStorageImageWriteWithoutFormat = true;
     features.textureCompressionASTC_LDR = is_optimal_astc_supported;
 
-    vk::PhysicalDeviceVertexAttributeDivisorFeaturesEXT vertex_divisor;
-    vertex_divisor.vertexAttributeInstanceRateDivisor = true;
-    vertex_divisor.vertexAttributeInstanceRateZeroDivisor = true;
-    SetNext(next, vertex_divisor);
+    vk::PhysicalDevice16BitStorageFeaturesKHR bit16_storage;
+    bit16_storage.uniformAndStorageBuffer16BitAccess = true;
+    SetNext(next, bit16_storage);
+
+    vk::PhysicalDevice8BitStorageFeaturesKHR bit8_storage;
+    bit8_storage.uniformAndStorageBuffer8BitAccess = true;
+    SetNext(next, bit8_storage);
 
     vk::PhysicalDeviceFloat16Int8FeaturesKHR float16_int8;
     if (is_float16_supported) {
@@ -120,6 +138,10 @@ bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instan
         LOG_INFO(Render_Vulkan, "Device doesn't support uint8 indexes");
     }
 
+    if (!ext_depth_range_unrestricted) {
+        LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted");
+    }
+
     vk::DeviceCreateInfo device_ci({}, static_cast<u32>(queue_cis.size()), queue_cis.data(), 0,
                                    nullptr, static_cast<u32>(extensions.size()), extensions.data(),
                                    nullptr);
@@ -135,16 +157,7 @@ bool VKDevice::Create(const vk::DispatchLoaderDynamic& dldi, vk::Instance instan
     logical = UniqueDevice(
         dummy_logical, vk::ObjectDestroy<vk::NoParent, vk::DispatchLoaderDynamic>(nullptr, dld));
 
-    if (khr_driver_properties) {
-        vk::PhysicalDeviceDriverPropertiesKHR driver;
-        vk::PhysicalDeviceProperties2 properties;
-        properties.pNext = &driver;
-        physical.getProperties2(&properties, dld);
-        driver_id = driver.driverID;
-        LOG_INFO(Render_Vulkan, "Driver: {} {}", driver.driverName, driver.driverInfo);
-    } else {
-        LOG_INFO(Render_Vulkan, "Driver: Unknown");
-    }
+    CollectTelemetryParameters();
 
     graphics_queue = logical->getQueue(graphics_family, 0, dld);
     present_queue = logical->getQueue(present_family, 0, dld);
@@ -190,6 +203,18 @@ vk::Format VKDevice::GetSupportedFormat(vk::Format wanted_format,
 
 bool VKDevice::IsOptimalAstcSupported(const vk::PhysicalDeviceFeatures& features,
                                       const vk::DispatchLoaderDynamic& dldi) const {
+    // Disable for now to avoid converting ASTC twice.
+    return false;
+    static constexpr std::array astc_formats = {
+        vk::Format::eAstc4x4SrgbBlock,    vk::Format::eAstc8x8SrgbBlock,
+        vk::Format::eAstc8x5SrgbBlock,    vk::Format::eAstc5x4SrgbBlock,
+        vk::Format::eAstc5x5UnormBlock,   vk::Format::eAstc5x5SrgbBlock,
+        vk::Format::eAstc10x8UnormBlock,  vk::Format::eAstc10x8SrgbBlock,
+        vk::Format::eAstc6x6UnormBlock,   vk::Format::eAstc6x6SrgbBlock,
+        vk::Format::eAstc10x10UnormBlock, vk::Format::eAstc10x10SrgbBlock,
+        vk::Format::eAstc12x12UnormBlock, vk::Format::eAstc12x12SrgbBlock,
+        vk::Format::eAstc8x6UnormBlock,   vk::Format::eAstc8x6SrgbBlock,
+        vk::Format::eAstc6x5UnormBlock,   vk::Format::eAstc6x5SrgbBlock};
     if (!features.textureCompressionASTC_LDR) {
         return false;
     }
@@ -197,12 +222,6 @@ bool VKDevice::IsOptimalAstcSupported(const vk::PhysicalDeviceFeatures& features
         vk::FormatFeatureFlagBits::eSampledImage | vk::FormatFeatureFlagBits::eBlitSrc |
         vk::FormatFeatureFlagBits::eBlitDst | vk::FormatFeatureFlagBits::eTransferSrc |
         vk::FormatFeatureFlagBits::eTransferDst};
-    constexpr std::array astc_formats = {
-        vk::Format::eAstc4x4UnormBlock, vk::Format::eAstc4x4SrgbBlock,
-        vk::Format::eAstc8x8SrgbBlock,  vk::Format::eAstc8x6SrgbBlock,
-        vk::Format::eAstc5x4SrgbBlock,  vk::Format::eAstc5x5UnormBlock,
-        vk::Format::eAstc5x5SrgbBlock,  vk::Format::eAstc10x8UnormBlock,
-        vk::Format::eAstc10x8SrgbBlock};
     for (const auto format : astc_formats) {
         const auto format_properties{physical.getFormatProperties(format, dldi)};
         if (!(format_properties.optimalTilingFeatures & format_feature_usage)) {
@@ -225,11 +244,17 @@ bool VKDevice::IsFormatSupported(vk::Format wanted_format, vk::FormatFeatureFlag
 
 bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical,
                           vk::SurfaceKHR surface) {
-    LOG_INFO(Render_Vulkan, "{}", physical.getProperties(dldi).deviceName);
     bool is_suitable = true;
 
-    constexpr std::array required_extensions = {VK_KHR_SWAPCHAIN_EXTENSION_NAME,
-                                                VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME};
+    constexpr std::array required_extensions = {
+        VK_KHR_SWAPCHAIN_EXTENSION_NAME,
+        VK_KHR_16BIT_STORAGE_EXTENSION_NAME,
+        VK_KHR_8BIT_STORAGE_EXTENSION_NAME,
+        VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME,
+        VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME,
+        VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME,
+        VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME,
+    };
     std::bitset<required_extensions.size()> available_extensions{};
 
     for (const auto& prop : physical.enumerateDeviceExtensionProperties(nullptr, dldi)) {
@@ -246,7 +271,7 @@ bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDev
             if (available_extensions[i]) {
                 continue;
             }
-            LOG_INFO(Render_Vulkan, "Missing required extension: {}", required_extensions[i]);
+            LOG_ERROR(Render_Vulkan, "Missing required extension: {}", required_extensions[i]);
             is_suitable = false;
         }
     }
@@ -263,7 +288,7 @@ bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDev
         has_present |= physical.getSurfaceSupportKHR(i, surface, dldi) != 0;
     }
     if (!has_graphics || !has_present) {
-        LOG_INFO(Render_Vulkan, "Device lacks a graphics and present queue");
+        LOG_ERROR(Render_Vulkan, "Device lacks a graphics and present queue");
         is_suitable = false;
     }
 
@@ -273,8 +298,15 @@ bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDev
 
     constexpr u32 required_ubo_size = 65536;
     if (limits.maxUniformBufferRange < required_ubo_size) {
-        LOG_INFO(Render_Vulkan, "Device UBO size {} is too small, {} is required)",
-                 limits.maxUniformBufferRange, required_ubo_size);
+        LOG_ERROR(Render_Vulkan, "Device UBO size {} is too small, {} is required",
+                  limits.maxUniformBufferRange, required_ubo_size);
+        is_suitable = false;
+    }
+
+    constexpr u32 required_num_viewports = 16;
+    if (limits.maxViewports < required_num_viewports) {
+        LOG_INFO(Render_Vulkan, "Device number of viewports {} is too small, {} is required",
+                 limits.maxViewports, required_num_viewports);
         is_suitable = false;
     }
 
@@ -285,24 +317,32 @@ bool VKDevice::IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDev
         std::make_pair(features.depthClamp, "depthClamp"),
         std::make_pair(features.samplerAnisotropy, "samplerAnisotropy"),
         std::make_pair(features.largePoints, "largePoints"),
+        std::make_pair(features.multiViewport, "multiViewport"),
+        std::make_pair(features.depthBiasClamp, "depthBiasClamp"),
+        std::make_pair(features.geometryShader, "geometryShader"),
+        std::make_pair(features.tessellationShader, "tessellationShader"),
+        std::make_pair(features.fragmentStoresAndAtomics, "fragmentStoresAndAtomics"),
+        std::make_pair(features.shaderImageGatherExtended, "shaderImageGatherExtended"),
+        std::make_pair(features.shaderStorageImageWriteWithoutFormat,
+                       "shaderStorageImageWriteWithoutFormat"),
     };
     for (const auto& [supported, name] : feature_report) {
         if (supported) {
             continue;
         }
-        LOG_INFO(Render_Vulkan, "Missing required feature: {}", name);
+        LOG_ERROR(Render_Vulkan, "Missing required feature: {}", name);
         is_suitable = false;
     }
 
+    if (!is_suitable) {
+        LOG_ERROR(Render_Vulkan, "{} is not suitable", properties.deviceName);
+    }
+
     return is_suitable;
 }
 
 std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynamic& dldi) {
     std::vector<const char*> extensions;
-    extensions.reserve(7);
-    extensions.push_back(VK_KHR_SWAPCHAIN_EXTENSION_NAME);
-    extensions.push_back(VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME);
-
     const auto Test = [&](const vk::ExtensionProperties& extension,
                           std::optional<std::reference_wrapper<bool>> status, const char* name,
                           bool push) {
@@ -317,13 +357,30 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami
         }
     };
 
+    extensions.reserve(13);
+    extensions.push_back(VK_KHR_SWAPCHAIN_EXTENSION_NAME);
+    extensions.push_back(VK_KHR_16BIT_STORAGE_EXTENSION_NAME);
+    extensions.push_back(VK_KHR_8BIT_STORAGE_EXTENSION_NAME);
+    extensions.push_back(VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME);
+    extensions.push_back(VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME);
+    extensions.push_back(VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME);
+    extensions.push_back(VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME);
+
+    [[maybe_unused]] const bool nsight =
+        std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED");
     bool khr_shader_float16_int8{};
+    bool ext_subgroup_size_control{};
     for (const auto& extension : physical.enumerateDeviceExtensionProperties(nullptr, dldi)) {
         Test(extension, khr_uniform_buffer_standard_layout,
              VK_KHR_UNIFORM_BUFFER_STANDARD_LAYOUT_EXTENSION_NAME, true);
-        Test(extension, ext_index_type_uint8, VK_EXT_INDEX_TYPE_UINT8_EXTENSION_NAME, true);
-        Test(extension, khr_driver_properties, VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME, true);
         Test(extension, khr_shader_float16_int8, VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME, false);
+        Test(extension, ext_depth_range_unrestricted,
+             VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME, true);
+        Test(extension, ext_index_type_uint8, VK_EXT_INDEX_TYPE_UINT8_EXTENSION_NAME, true);
+        Test(extension, ext_shader_viewport_index_layer,
+             VK_EXT_SHADER_VIEWPORT_INDEX_LAYER_EXTENSION_NAME, true);
+        Test(extension, ext_subgroup_size_control, VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME,
+             false);
     }
 
     if (khr_shader_float16_int8) {
@@ -332,6 +389,23 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami
         extensions.push_back(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME);
     }
 
+    if (ext_subgroup_size_control) {
+        const auto features =
+            GetFeatures<vk::PhysicalDeviceSubgroupSizeControlFeaturesEXT>(physical, dldi);
+        const auto properties =
+            GetProperties<vk::PhysicalDeviceSubgroupSizeControlPropertiesEXT>(physical, dldi);
+
+        is_warp_potentially_bigger = properties.maxSubgroupSize > GuestWarpSize;
+
+        if (features.subgroupSizeControl && properties.minSubgroupSize <= GuestWarpSize &&
+            properties.maxSubgroupSize >= GuestWarpSize) {
+            extensions.push_back(VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME);
+            guest_warp_stages = properties.requiredSubgroupSizeStages;
+        }
+    } else {
+        is_warp_potentially_bigger = true;
+    }
+
     return extensions;
 }
 
@@ -358,19 +432,23 @@ void VKDevice::SetupFamilies(const vk::DispatchLoaderDynamic& dldi, vk::SurfaceK
     present_family = *present_family_;
 }
 
-void VKDevice::SetupProperties(const vk::DispatchLoaderDynamic& dldi) {
-    const auto props = physical.getProperties(dldi);
-    device_type = props.deviceType;
-    uniform_buffer_alignment = static_cast<u64>(props.limits.minUniformBufferOffsetAlignment);
-    storage_buffer_alignment = static_cast<u64>(props.limits.minStorageBufferOffsetAlignment);
-    max_storage_buffer_range = static_cast<u64>(props.limits.maxStorageBufferRange);
-}
-
 void VKDevice::SetupFeatures(const vk::DispatchLoaderDynamic& dldi) {
     const auto supported_features{physical.getFeatures(dldi)};
     is_optimal_astc_supported = IsOptimalAstcSupported(supported_features, dldi);
 }
 
+void VKDevice::CollectTelemetryParameters() {
+    const auto driver = GetProperties<vk::PhysicalDeviceDriverPropertiesKHR>(physical, dld);
+    driver_id = driver.driverID;
+    vendor_name = driver.driverName;
+
+    const auto extensions = physical.enumerateDeviceExtensionProperties(nullptr, dld);
+    reported_extensions.reserve(std::size(extensions));
+    for (const auto& extension : extensions) {
+        reported_extensions.push_back(extension.extensionName);
+    }
+}
+
 std::vector<vk::DeviceQueueCreateInfo> VKDevice::GetDeviceQueueCreateInfos() const {
     static const float QUEUE_PRIORITY = 1.0f;
 
@@ -385,50 +463,70 @@ std::vector<vk::DeviceQueueCreateInfo> VKDevice::GetDeviceQueueCreateInfos() con
 
 std::unordered_map<vk::Format, vk::FormatProperties> VKDevice::GetFormatProperties(
     const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical) {
-    constexpr std::array formats{vk::Format::eA8B8G8R8UnormPack32,
-                                 vk::Format::eA8B8G8R8SnormPack32,
-                                 vk::Format::eA8B8G8R8SrgbPack32,
-                                 vk::Format::eB5G6R5UnormPack16,
-                                 vk::Format::eA2B10G10R10UnormPack32,
-                                 vk::Format::eR32G32B32A32Sfloat,
-                                 vk::Format::eR16G16B16A16Uint,
-                                 vk::Format::eR16G16Unorm,
-                                 vk::Format::eR16G16Snorm,
-                                 vk::Format::eR16G16Sfloat,
-                                 vk::Format::eR16Unorm,
-                                 vk::Format::eR8G8B8A8Srgb,
-                                 vk::Format::eR8G8Unorm,
-                                 vk::Format::eR8G8Snorm,
-                                 vk::Format::eR8Unorm,
-                                 vk::Format::eB10G11R11UfloatPack32,
-                                 vk::Format::eR32Sfloat,
-                                 vk::Format::eR16Sfloat,
-                                 vk::Format::eR16G16B16A16Sfloat,
-                                 vk::Format::eB8G8R8A8Unorm,
-                                 vk::Format::eD32Sfloat,
-                                 vk::Format::eD16Unorm,
-                                 vk::Format::eD16UnormS8Uint,
-                                 vk::Format::eD24UnormS8Uint,
-                                 vk::Format::eD32SfloatS8Uint,
-                                 vk::Format::eBc1RgbaUnormBlock,
-                                 vk::Format::eBc2UnormBlock,
-                                 vk::Format::eBc3UnormBlock,
-                                 vk::Format::eBc4UnormBlock,
-                                 vk::Format::eBc5UnormBlock,
-                                 vk::Format::eBc5SnormBlock,
-                                 vk::Format::eBc7UnormBlock,
-                                 vk::Format::eBc1RgbaSrgbBlock,
-                                 vk::Format::eBc3SrgbBlock,
-                                 vk::Format::eBc7SrgbBlock,
-                                 vk::Format::eAstc4x4UnormBlock,
-                                 vk::Format::eAstc4x4SrgbBlock,
-                                 vk::Format::eAstc8x8SrgbBlock,
-                                 vk::Format::eAstc8x6SrgbBlock,
-                                 vk::Format::eAstc5x4SrgbBlock,
-                                 vk::Format::eAstc5x5UnormBlock,
-                                 vk::Format::eAstc5x5SrgbBlock,
-                                 vk::Format::eAstc10x8UnormBlock,
-                                 vk::Format::eAstc10x8SrgbBlock};
+    static constexpr std::array formats{vk::Format::eA8B8G8R8UnormPack32,
+                                        vk::Format::eA8B8G8R8SnormPack32,
+                                        vk::Format::eA8B8G8R8SrgbPack32,
+                                        vk::Format::eB5G6R5UnormPack16,
+                                        vk::Format::eA2B10G10R10UnormPack32,
+                                        vk::Format::eA1R5G5B5UnormPack16,
+                                        vk::Format::eR32G32B32A32Sfloat,
+                                        vk::Format::eR32G32B32A32Uint,
+                                        vk::Format::eR32G32Sfloat,
+                                        vk::Format::eR32G32Uint,
+                                        vk::Format::eR16G16B16A16Uint,
+                                        vk::Format::eR16G16B16A16Unorm,
+                                        vk::Format::eR16G16Unorm,
+                                        vk::Format::eR16G16Snorm,
+                                        vk::Format::eR16G16Sfloat,
+                                        vk::Format::eR16Unorm,
+                                        vk::Format::eR8G8B8A8Srgb,
+                                        vk::Format::eR8G8Unorm,
+                                        vk::Format::eR8G8Snorm,
+                                        vk::Format::eR8Unorm,
+                                        vk::Format::eR8Uint,
+                                        vk::Format::eB10G11R11UfloatPack32,
+                                        vk::Format::eR32Sfloat,
+                                        vk::Format::eR32Uint,
+                                        vk::Format::eR16Sfloat,
+                                        vk::Format::eR16G16B16A16Sfloat,
+                                        vk::Format::eB8G8R8A8Unorm,
+                                        vk::Format::eR4G4B4A4UnormPack16,
+                                        vk::Format::eD32Sfloat,
+                                        vk::Format::eD16Unorm,
+                                        vk::Format::eD16UnormS8Uint,
+                                        vk::Format::eD24UnormS8Uint,
+                                        vk::Format::eD32SfloatS8Uint,
+                                        vk::Format::eBc1RgbaUnormBlock,
+                                        vk::Format::eBc2UnormBlock,
+                                        vk::Format::eBc3UnormBlock,
+                                        vk::Format::eBc4UnormBlock,
+                                        vk::Format::eBc5UnormBlock,
+                                        vk::Format::eBc5SnormBlock,
+                                        vk::Format::eBc7UnormBlock,
+                                        vk::Format::eBc6HUfloatBlock,
+                                        vk::Format::eBc6HSfloatBlock,
+                                        vk::Format::eBc1RgbaSrgbBlock,
+                                        vk::Format::eBc3SrgbBlock,
+                                        vk::Format::eBc7SrgbBlock,
+                                        vk::Format::eAstc4x4SrgbBlock,
+                                        vk::Format::eAstc8x8SrgbBlock,
+                                        vk::Format::eAstc8x5SrgbBlock,
+                                        vk::Format::eAstc5x4SrgbBlock,
+                                        vk::Format::eAstc5x5UnormBlock,
+                                        vk::Format::eAstc5x5SrgbBlock,
+                                        vk::Format::eAstc10x8UnormBlock,
+                                        vk::Format::eAstc10x8SrgbBlock,
+                                        vk::Format::eAstc6x6UnormBlock,
+                                        vk::Format::eAstc6x6SrgbBlock,
+                                        vk::Format::eAstc10x10UnormBlock,
+                                        vk::Format::eAstc10x10SrgbBlock,
+                                        vk::Format::eAstc12x12UnormBlock,
+                                        vk::Format::eAstc12x12SrgbBlock,
+                                        vk::Format::eAstc8x6UnormBlock,
+                                        vk::Format::eAstc8x6SrgbBlock,
+                                        vk::Format::eAstc6x5UnormBlock,
+                                        vk::Format::eAstc6x5SrgbBlock,
+                                        vk::Format::eE5B9G9R9UfloatPack32};
     std::unordered_map<vk::Format, vk::FormatProperties> format_properties;
     for (const auto format : formats) {
         format_properties.emplace(format, physical.getFormatProperties(format, dldi));

src/video_core/renderer_vulkan/vk_device.h

@@ -4,6 +4,8 @@
 
 #pragma once
 
+#include <string>
+#include <string_view>
 #include <unordered_map>
 #include <vector>
 #include "common/common_types.h"
@@ -14,6 +16,9 @@ namespace Vulkan {
 /// Format usage descriptor.
 enum class FormatType { Linear, Optimal, Buffer };
 
+/// Subgroup size of the guest emulated hardware (Nvidia has 32 threads per subgroup).
+const u32 GuestWarpSize = 32;
+
 /// Handles data specific to a physical device.
 class VKDevice final {
 public:
@@ -71,7 +76,22 @@ public:
 
     /// Returns true if the device is integrated with the host CPU.
     bool IsIntegrated() const {
-        return device_type == vk::PhysicalDeviceType::eIntegratedGpu;
+        return properties.deviceType == vk::PhysicalDeviceType::eIntegratedGpu;
+    }
+
+    /// Returns the current Vulkan API version provided in Vulkan-formatted version numbers.
+    u32 GetApiVersion() const {
+        return properties.apiVersion;
+    }
+
+    /// Returns the current driver version provided in Vulkan-formatted version numbers.
+    u32 GetDriverVersion() const {
+        return properties.driverVersion;
+    }
+
+    /// Returns the device name.
+    std::string_view GetModelName() const {
+        return properties.deviceName;
     }
 
     /// Returns the driver ID.
@@ -80,18 +100,23 @@ public:
     }
 
     /// Returns uniform buffer alignment requeriment.
-    u64 GetUniformBufferAlignment() const {
-        return uniform_buffer_alignment;
+    vk::DeviceSize GetUniformBufferAlignment() const {
+        return properties.limits.minUniformBufferOffsetAlignment;
     }
 
     /// Returns storage alignment requeriment.
-    u64 GetStorageBufferAlignment() const {
-        return storage_buffer_alignment;
+    vk::DeviceSize GetStorageBufferAlignment() const {
+        return properties.limits.minStorageBufferOffsetAlignment;
     }
 
     /// Returns the maximum range for storage buffers.
-    u64 GetMaxStorageBufferRange() const {
-        return max_storage_buffer_range;
+    vk::DeviceSize GetMaxStorageBufferRange() const {
+        return properties.limits.maxStorageBufferRange;
+    }
+
+    /// Returns the maximum size for push constants.
+    vk::DeviceSize GetMaxPushConstantsSize() const {
+        return properties.limits.maxPushConstantsSize;
     }
 
     /// Returns true if ASTC is natively supported.
@@ -104,6 +129,16 @@ public:
         return is_float16_supported;
     }
 
+    /// Returns true if the device warp size can potentially be bigger than guest's warp size.
+    bool IsWarpSizePotentiallyBiggerThanGuest() const {
+        return is_warp_potentially_bigger;
+    }
+
+    /// Returns true if the device can be forced to use the guest warp size.
+    bool IsGuestWarpSizeSupported(vk::ShaderStageFlagBits stage) const {
+        return (guest_warp_stages & stage) != vk::ShaderStageFlags{};
+    }
+
     /// Returns true if the device supports VK_EXT_scalar_block_layout.
     bool IsKhrUniformBufferStandardLayoutSupported() const {
         return khr_uniform_buffer_standard_layout;
@@ -114,6 +149,26 @@ public:
         return ext_index_type_uint8;
     }
 
+    /// Returns true if the device supports VK_EXT_depth_range_unrestricted.
+    bool IsExtDepthRangeUnrestrictedSupported() const {
+        return ext_depth_range_unrestricted;
+    }
+
+    /// Returns true if the device supports VK_EXT_shader_viewport_index_layer.
+    bool IsExtShaderViewportIndexLayerSupported() const {
+        return ext_shader_viewport_index_layer;
+    }
+
+    /// Returns the vendor name reported from Vulkan.
+    std::string_view GetVendorName() const {
+        return vendor_name;
+    }
+
+    /// Returns the list of available extensions.
+    const std::vector<std::string>& GetAvailableExtensions() const {
+        return reported_extensions;
+    }
+
     /// Checks if the physical device is suitable.
     static bool IsSuitable(const vk::DispatchLoaderDynamic& dldi, vk::PhysicalDevice physical,
                            vk::SurfaceKHR surface);
@@ -125,12 +180,12 @@ private:
     /// Sets up queue families.
     void SetupFamilies(const vk::DispatchLoaderDynamic& dldi, vk::SurfaceKHR surface);
 
-    /// Sets up device properties.
-    void SetupProperties(const vk::DispatchLoaderDynamic& dldi);
-
     /// Sets up device features.
     void SetupFeatures(const vk::DispatchLoaderDynamic& dldi);
 
+    /// Collects telemetry information from the device.
+    void CollectTelemetryParameters();
+
     /// Returns a list of queue initialization descriptors.
     std::vector<vk::DeviceQueueCreateInfo> GetDeviceQueueCreateInfos() const;
 
@@ -148,23 +203,28 @@ private:
 
     const vk::PhysicalDevice physical;         ///< Physical device.
     vk::DispatchLoaderDynamic dld;             ///< Device function pointers.
+    vk::PhysicalDeviceProperties properties;   ///< Device properties.
     UniqueDevice logical;                      ///< Logical device.
     vk::Queue graphics_queue;                  ///< Main graphics queue.
     vk::Queue present_queue;                   ///< Main present queue.
     u32 graphics_family{};                     ///< Main graphics queue family index.
     u32 present_family{};                      ///< Main present queue family index.
-    vk::PhysicalDeviceType device_type;        ///< Physical device type.
     vk::DriverIdKHR driver_id{};               ///< Driver ID.
-    u64 uniform_buffer_alignment{};            ///< Uniform buffer alignment requeriment.
-    u64 storage_buffer_alignment{};            ///< Storage buffer alignment requeriment.
-    u64 max_storage_buffer_range{};            ///< Max storage buffer size.
+    vk::ShaderStageFlags guest_warp_stages{};  ///< Stages where the guest warp size can be forced.
     bool is_optimal_astc_supported{};          ///< Support for native ASTC.
     bool is_float16_supported{};               ///< Support for float16 arithmetics.
+    bool is_warp_potentially_bigger{};         ///< Host warp size can be bigger than guest.
     bool khr_uniform_buffer_standard_layout{}; ///< Support for std430 on UBOs.
     bool ext_index_type_uint8{};               ///< Support for VK_EXT_index_type_uint8.
-    bool khr_driver_properties{};              ///< Support for VK_KHR_driver_properties.
-    std::unordered_map<vk::Format, vk::FormatProperties>
-        format_properties; ///< Format properties dictionary.
+    bool ext_depth_range_unrestricted{};       ///< Support for VK_EXT_depth_range_unrestricted.
+    bool ext_shader_viewport_index_layer{};    ///< Support for VK_EXT_shader_viewport_index_layer.
+
+    // Telemetry parameters
+    std::string vendor_name;                      ///< Device's driver name.
+    std::vector<std::string> reported_extensions; ///< Reported Vulkan extensions.
+
+    /// Format properties dictionary.
+    std::unordered_map<vk::Format, vk::FormatProperties> format_properties;
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -983,6 +983,11 @@ private:
         return {};
     }
 
+    Id TextureGradient(Operation operation) {
+        UNIMPLEMENTED();
+        return {};
+    }
+
     Id ImageLoad(Operation operation) {
         UNIMPLEMENTED();
         return {};
@@ -1391,6 +1396,7 @@ private:
         &SPIRVDecompiler::Quaternary<&Module::OpBitFieldInsert, Type::Int>,
         &SPIRVDecompiler::Ternary<&Module::OpBitFieldSExtract, Type::Int>,
         &SPIRVDecompiler::Unary<&Module::OpBitCount, Type::Int>,
+        &SPIRVDecompiler::Unary<&Module::OpFindSMsb, Type::Int>,
 
         &SPIRVDecompiler::Binary<&Module::OpIAdd, Type::Uint>,
         &SPIRVDecompiler::Binary<&Module::OpIMul, Type::Uint>,
@@ -1409,6 +1415,7 @@ private:
         &SPIRVDecompiler::Quaternary<&Module::OpBitFieldInsert, Type::Uint>,
         &SPIRVDecompiler::Ternary<&Module::OpBitFieldUExtract, Type::Uint>,
         &SPIRVDecompiler::Unary<&Module::OpBitCount, Type::Uint>,
+        &SPIRVDecompiler::Unary<&Module::OpFindUMsb, Type::Uint>,
 
         &SPIRVDecompiler::Binary<&Module::OpFAdd, Type::HalfFloat>,
         &SPIRVDecompiler::Binary<&Module::OpFMul, Type::HalfFloat>,
@@ -1473,6 +1480,7 @@ private:
         &SPIRVDecompiler::TextureQueryDimensions,
         &SPIRVDecompiler::TextureQueryLod,
         &SPIRVDecompiler::TexelFetch,
+        &SPIRVDecompiler::TextureGradient,
 
         &SPIRVDecompiler::ImageLoad,
         &SPIRVDecompiler::ImageStore,

src/video_core/renderer_vulkan/vk_swapchain.cpp

@@ -19,12 +19,18 @@
 namespace Vulkan {
 
 namespace {
-vk::SurfaceFormatKHR ChooseSwapSurfaceFormat(const std::vector<vk::SurfaceFormatKHR>& formats) {
+
+vk::SurfaceFormatKHR ChooseSwapSurfaceFormat(const std::vector<vk::SurfaceFormatKHR>& formats,
+                                             bool srgb) {
     if (formats.size() == 1 && formats[0].format == vk::Format::eUndefined) {
-        return {vk::Format::eB8G8R8A8Unorm, vk::ColorSpaceKHR::eSrgbNonlinear};
+        vk::SurfaceFormatKHR format;
+        format.format = vk::Format::eB8G8R8A8Unorm;
+        format.colorSpace = vk::ColorSpaceKHR::eSrgbNonlinear;
+        return format;
     }
-    const auto& found = std::find_if(formats.begin(), formats.end(), [](const auto& format) {
-        return format.format == vk::Format::eB8G8R8A8Unorm &&
+    const auto& found = std::find_if(formats.begin(), formats.end(), [srgb](const auto& format) {
+        const auto request_format = srgb ? vk::Format::eB8G8R8A8Srgb : vk::Format::eB8G8R8A8Unorm;
+        return format.format == request_format &&
                format.colorSpace == vk::ColorSpaceKHR::eSrgbNonlinear;
     });
     return found != formats.end() ? *found : formats[0];
@@ -51,28 +57,26 @@ vk::Extent2D ChooseSwapExtent(const vk::SurfaceCapabilitiesKHR& capabilities, u3
                              std::min(capabilities.maxImageExtent.height, extent.height));
     return extent;
 }
-} // namespace
+
+} // Anonymous namespace
 
 VKSwapchain::VKSwapchain(vk::SurfaceKHR surface, const VKDevice& device)
     : surface{surface}, device{device} {}
 
 VKSwapchain::~VKSwapchain() = default;
 
-void VKSwapchain::Create(u32 width, u32 height) {
-    const auto dev = device.GetLogical();
+void VKSwapchain::Create(u32 width, u32 height, bool srgb) {
     const auto& dld = device.GetDispatchLoader();
     const auto physical_device = device.GetPhysical();
-
-    const vk::SurfaceCapabilitiesKHR capabilities{
-        physical_device.getSurfaceCapabilitiesKHR(surface, dld)};
+    const auto capabilities{physical_device.getSurfaceCapabilitiesKHR(surface, dld)};
     if (capabilities.maxImageExtent.width == 0 || capabilities.maxImageExtent.height == 0) {
         return;
     }
 
-    dev.waitIdle(dld);
+    device.GetLogical().waitIdle(dld);
     Destroy();
 
-    CreateSwapchain(capabilities, width, height);
+    CreateSwapchain(capabilities, width, height, srgb);
     CreateSemaphores();
     CreateImageViews();
 
@@ -107,7 +111,7 @@ bool VKSwapchain::Present(vk::Semaphore render_semaphore, VKFence& fence) {
         break;
     case vk::Result::eErrorOutOfDateKHR:
         if (current_width > 0 && current_height > 0) {
-            Create(current_width, current_height);
+            Create(current_width, current_height, current_srgb);
             recreated = true;
         }
         break;
@@ -129,23 +133,19 @@ bool VKSwapchain::HasFramebufferChanged(const Layout::FramebufferLayout& framebu
 }
 
 void VKSwapchain::CreateSwapchain(const vk::SurfaceCapabilitiesKHR& capabilities, u32 width,
-                                  u32 height) {
-    const auto dev{device.GetLogical()};
+                                  u32 height, bool srgb) {
     const auto& dld{device.GetDispatchLoader()};
     const auto physical_device{device.GetPhysical()};
+    const auto formats{physical_device.getSurfaceFormatsKHR(surface, dld)};
+    const auto present_modes{physical_device.getSurfacePresentModesKHR(surface, dld)};
 
-    const std::vector<vk::SurfaceFormatKHR> formats{
-        physical_device.getSurfaceFormatsKHR(surface, dld)};
-
-    const std::vector<vk::PresentModeKHR> present_modes{
-        physical_device.getSurfacePresentModesKHR(surface, dld)};
-
-    const vk::SurfaceFormatKHR surface_format{ChooseSwapSurfaceFormat(formats)};
+    const vk::SurfaceFormatKHR surface_format{ChooseSwapSurfaceFormat(formats, srgb)};
     const vk::PresentModeKHR present_mode{ChooseSwapPresentMode(present_modes)};
     extent = ChooseSwapExtent(capabilities, width, height);
 
     current_width = extent.width;
     current_height = extent.height;
+    current_srgb = srgb;
 
     u32 requested_image_count{capabilities.minImageCount + 1};
     if (capabilities.maxImageCount > 0 && requested_image_count > capabilities.maxImageCount) {
@@ -169,6 +169,7 @@ void VKSwapchain::CreateSwapchain(const vk::SurfaceCapabilitiesKHR& capabilities
         swapchain_ci.imageSharingMode = vk::SharingMode::eExclusive;
     }
 
+    const auto dev{device.GetLogical()};
     swapchain = dev.createSwapchainKHRUnique(swapchain_ci, nullptr, dld);
 
     images = dev.getSwapchainImagesKHR(*swapchain, dld);

src/video_core/renderer_vulkan/vk_swapchain.h

@@ -24,7 +24,7 @@ public:
     ~VKSwapchain();
 
     /// Creates (or recreates) the swapchain with a given size.
-    void Create(u32 width, u32 height);
+    void Create(u32 width, u32 height, bool srgb);
 
     /// Acquires the next image in the swapchain, waits as needed.
     void AcquireNextImage();
@@ -60,8 +60,13 @@ public:
         return image_format;
     }
 
+    bool GetSrgbState() const {
+        return current_srgb;
+    }
+
 private:
-    void CreateSwapchain(const vk::SurfaceCapabilitiesKHR& capabilities, u32 width, u32 height);
+    void CreateSwapchain(const vk::SurfaceCapabilitiesKHR& capabilities, u32 width, u32 height,
+                         bool srgb);
     void CreateSemaphores();
     void CreateImageViews();
 
@@ -87,6 +92,7 @@ private:
 
     u32 current_width{};
     u32 current_height{};
+    bool current_srgb{};
 };
 
 } // namespace Vulkan

src/video_core/shader/decode/arithmetic_integer.cpp

@@ -130,6 +130,25 @@ u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) {
         SetRegister(bb, instr.gpr0, value);
         break;
     }
+    case OpCode::Id::FLO_R:
+    case OpCode::Id::FLO_C:
+    case OpCode::Id::FLO_IMM: {
+        Node value;
+        if (instr.flo.invert) {
+            op_b = Operation(OperationCode::IBitwiseNot, NO_PRECISE, std::move(op_b));
+        }
+        if (instr.flo.is_signed) {
+            value = Operation(OperationCode::IBitMSB, NO_PRECISE, std::move(op_b));
+        } else {
+            value = Operation(OperationCode::UBitMSB, NO_PRECISE, std::move(op_b));
+        }
+        if (instr.flo.sh) {
+            value =
+                Operation(OperationCode::UBitwiseXor, NO_PRECISE, std::move(value), Immediate(31));
+        }
+        SetRegister(bb, instr.gpr0, std::move(value));
+        break;
+    }
     case OpCode::Id::SEL_C:
     case OpCode::Id::SEL_R:
     case OpCode::Id::SEL_IMM: {

src/video_core/shader/decode/texture.cpp

@@ -134,13 +134,52 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
         Node4 values;
         for (u32 element = 0; element < values.size(); ++element) {
             auto coords_copy = coords;
-            MetaTexture meta{sampler, {}, {}, {}, {}, {}, component, element};
+            MetaTexture meta{sampler, {}, {}, {}, {}, {}, {}, component, element};
             values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy));
         }
 
         WriteTexsInstructionFloat(bb, instr, values, true);
         break;
     }
+    case OpCode::Id::TXD_B:
+        is_bindless = true;
+        [[fallthrough]];
+    case OpCode::Id::TXD: {
+        UNIMPLEMENTED_IF_MSG(instr.txd.UsesMiscMode(TextureMiscMode::AOFFI),
+                             "AOFFI is not implemented");
+        UNIMPLEMENTED_IF_MSG(instr.txd.is_array != 0, "TXD Array is not implemented");
+
+        u64 base_reg = instr.gpr8.Value();
+        const auto derivate_reg = instr.gpr20.Value();
+        const auto texture_type = instr.txd.texture_type.Value();
+        const auto coord_count = GetCoordCount(texture_type);
+
+        const auto& sampler = is_bindless
+                                  ? GetBindlessSampler(base_reg, {{texture_type, false, false}})
+                                  : GetSampler(instr.sampler, {{texture_type, false, false}});
+        if (is_bindless) {
+            base_reg++;
+        }
+
+        std::vector<Node> coords;
+        std::vector<Node> derivates;
+        for (std::size_t i = 0; i < coord_count; ++i) {
+            coords.push_back(GetRegister(base_reg + i));
+            const std::size_t derivate = i * 2;
+            derivates.push_back(GetRegister(derivate_reg + derivate));
+            derivates.push_back(GetRegister(derivate_reg + derivate + 1));
+        }
+
+        Node4 values;
+        for (u32 element = 0; element < values.size(); ++element) {
+            MetaTexture meta{sampler, {}, {}, {}, derivates, {}, {}, {}, element};
+            values[element] = Operation(OperationCode::TextureGradient, std::move(meta), coords);
+        }
+
+        WriteTexInstructionFloat(bb, instr, values);
+
+        break;
+    }
     case OpCode::Id::TXQ_B:
         is_bindless = true;
         [[fallthrough]];
@@ -158,7 +197,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
                 if (!instr.txq.IsComponentEnabled(element)) {
                     continue;
                 }
-                MetaTexture meta{sampler, {}, {}, {}, {}, {}, {}, element};
+                MetaTexture meta{sampler, {}, {}, {}, {}, {}, {}, {}, element};
                 const Node value =
                     Operation(OperationCode::TextureQueryDimensions, meta,
                               GetRegister(instr.gpr8.Value() + (is_bindless ? 1 : 0)));
@@ -212,7 +251,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
                 continue;
             }
             auto params = coords;
-            MetaTexture meta{sampler, {}, {}, {}, {}, {}, {}, element};
+            MetaTexture meta{sampler, {}, {}, {}, {}, {}, {}, {}, element};
             const Node value = Operation(OperationCode::TextureQueryLod, meta, std::move(params));
             SetTemporary(bb, indexer++, value);
         }
@@ -442,7 +481,7 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
     Node4 values;
     for (u32 element = 0; element < values.size(); ++element) {
         auto copy_coords = coords;
-        MetaTexture meta{sampler, array, depth_compare, aoffi, bias, lod, {}, element};
+        MetaTexture meta{sampler, array, depth_compare, aoffi, {}, bias, lod, {}, element};
         values[element] = Operation(read_method, meta, std::move(copy_coords));
     }
 
@@ -574,7 +613,7 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de
     Node4 values;
     for (u32 element = 0; element < values.size(); ++element) {
         auto coords_copy = coords;
-        MetaTexture meta{sampler, GetRegister(array_register), dc, aoffi, {}, {}, component,
+        MetaTexture meta{sampler, GetRegister(array_register), dc, aoffi, {}, {}, {}, component,
                          element};
         values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy));
     }
@@ -608,7 +647,7 @@ Node4 ShaderIR::GetTldCode(Tegra::Shader::Instruction instr) {
     Node4 values;
     for (u32 element = 0; element < values.size(); ++element) {
         auto coords_copy = coords;
-        MetaTexture meta{sampler, array_register, {}, {}, {}, lod, {}, element};
+        MetaTexture meta{sampler, array_register, {}, {}, {}, {}, lod, {}, element};
         values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy));
     }
 
@@ -653,7 +692,7 @@ Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is
     Node4 values;
     for (u32 element = 0; element < values.size(); ++element) {
         auto coords_copy = coords;
-        MetaTexture meta{sampler, array, {}, {}, {}, lod, {}, element};
+        MetaTexture meta{sampler, array, {}, {}, {}, {}, lod, {}, element};
         values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy));
     }
     return values;

src/video_core/shader/node.h

@@ -68,6 +68,7 @@ enum class OperationCode {
     IBitfieldInsert,       /// (MetaArithmetic, int base, int insert, int offset, int bits) -> int
     IBitfieldExtract,      /// (MetaArithmetic, int value, int offset, int offset) -> int
     IBitCount,             /// (MetaArithmetic, int) -> int
+    IBitMSB,               /// (MetaArithmetic, int) -> int
 
     UAdd,                  /// (MetaArithmetic, uint a, uint b) -> uint
     UMul,                  /// (MetaArithmetic, uint a, uint b) -> uint
@@ -86,6 +87,7 @@ enum class OperationCode {
     UBitfieldInsert,  /// (MetaArithmetic, uint base, uint insert, int offset, int bits) -> uint
     UBitfieldExtract, /// (MetaArithmetic, uint value, int offset, int offset) -> uint
     UBitCount,        /// (MetaArithmetic, uint) -> uint
+    UBitMSB,          /// (MetaArithmetic, uint) -> uint
 
     HAdd,       /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
     HMul,       /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
@@ -149,6 +151,7 @@ enum class OperationCode {
     TextureQueryDimensions, /// (MetaTexture, float a) -> float4
     TextureQueryLod,        /// (MetaTexture, float[N] coords) -> float4
     TexelFetch,             /// (MetaTexture, int[N], int) -> float4
+    TextureGradient,        /// (MetaTexture, float[N] coords, float[N*2] derivates) -> float4
 
     ImageLoad,  /// (MetaImage, int[N] coords) -> void
     ImageStore, /// (MetaImage, int[N] coords) -> void
@@ -367,6 +370,7 @@ struct MetaTexture {
     Node array;
     Node depth_compare;
     std::vector<Node> aoffi;
+    std::vector<Node> derivates;
     Node bias;
     Node lod;
     Node component{};

src/video_core/texture_cache/surface_base.h

@@ -254,16 +254,14 @@ public:
         if (!layer_mipmap) {
             return {};
         }
-        const u32 end_layer{layer_mipmap->first};
-        const u32 end_mipmap{layer_mipmap->second};
+        const auto [end_layer, end_mipmap] = *layer_mipmap;
         if (layer != end_layer) {
             if (mipmap == 0 && end_mipmap == 0) {
-                return GetView(ViewParams(view_params.target, layer, end_layer - layer + 1, 0, 1));
+                return GetView(ViewParams(view_params.target, layer, end_layer - layer, 0, 1));
             }
             return {};
         } else {
-            return GetView(
-                ViewParams(view_params.target, layer, 1, mipmap, end_mipmap - mipmap + 1));
+            return GetView(ViewParams(view_params.target, layer, 1, mipmap, end_mipmap - mipmap));
         }
     }
 
@@ -278,8 +276,7 @@ public:
         if (!layer_mipmap) {
             return {};
         }
-        const u32 layer{layer_mipmap->first};
-        const u32 mipmap{layer_mipmap->second};
+        const auto [layer, mipmap] = *layer_mipmap;
         if (GetMipmapSize(mipmap) != candidate_size) {
             return EmplaceIrregularView(view_params, view_addr, candidate_size, mipmap, layer);
         }

src/video_core/texture_cache/surface_params.cpp

@@ -246,6 +246,16 @@ SurfaceParams SurfaceParams::CreateForFermiCopySurface(
     return params;
 }
 
+VideoCore::Surface::SurfaceTarget SurfaceParams::ExpectedTarget(
+    const VideoCommon::Shader::Sampler& entry) {
+    return TextureTypeToSurfaceTarget(entry.GetType(), entry.IsArray());
+}
+
+VideoCore::Surface::SurfaceTarget SurfaceParams::ExpectedTarget(
+    const VideoCommon::Shader::Image& entry) {
+    return ImageTypeToSurfaceTarget(entry.GetType());
+}
+
 bool SurfaceParams::IsLayered() const {
     switch (target) {
     case SurfaceTarget::Texture1DArray:

src/video_core/texture_cache/surface_params.h

@@ -45,6 +45,14 @@ public:
     static SurfaceParams CreateForFermiCopySurface(
         const Tegra::Engines::Fermi2D::Regs::Surface& config);
 
+    /// Obtains the texture target from a shader's sampler entry.
+    static VideoCore::Surface::SurfaceTarget ExpectedTarget(
+        const VideoCommon::Shader::Sampler& entry);
+
+    /// Obtains the texture target from a shader's sampler entry.
+    static VideoCore::Surface::SurfaceTarget ExpectedTarget(
+        const VideoCommon::Shader::Image& entry);
+
     std::size_t Hash() const {
         return static_cast<std::size_t>(
             Common::CityHash64(reinterpret_cast<const char*>(this), sizeof(*this)));

src/video_core/texture_cache/texture_cache.h

@@ -95,10 +95,16 @@ public:
         std::lock_guard lock{mutex};
         const auto gpu_addr{tic.Address()};
         if (!gpu_addr) {
-            return {};
+            return GetNullSurface(SurfaceParams::ExpectedTarget(entry));
+        }
+
+        const auto host_ptr{system.GPU().MemoryManager().GetPointer(gpu_addr)};
+        const auto cache_addr{ToCacheAddr(host_ptr)};
+        if (!cache_addr) {
+            return GetNullSurface(SurfaceParams::ExpectedTarget(entry));
         }
         const auto params{SurfaceParams::CreateForTexture(format_lookup_table, tic, entry)};
-        const auto [surface, view] = GetSurface(gpu_addr, params, true, false);
+        const auto [surface, view] = GetSurface(gpu_addr, cache_addr, params, true, false);
         if (guard_samplers) {
             sampled_textures.push_back(surface);
         }
@@ -110,10 +116,15 @@ public:
         std::lock_guard lock{mutex};
         const auto gpu_addr{tic.Address()};
         if (!gpu_addr) {
-            return {};
+            return GetNullSurface(SurfaceParams::ExpectedTarget(entry));
+        }
+        const auto host_ptr{system.GPU().MemoryManager().GetPointer(gpu_addr)};
+        const auto cache_addr{ToCacheAddr(host_ptr)};
+        if (!cache_addr) {
+            return GetNullSurface(SurfaceParams::ExpectedTarget(entry));
         }
         const auto params{SurfaceParams::CreateForImage(format_lookup_table, tic, entry)};
-        const auto [surface, view] = GetSurface(gpu_addr, params, true, false);
+        const auto [surface, view] = GetSurface(gpu_addr, cache_addr, params, true, false);
         if (guard_samplers) {
             sampled_textures.push_back(surface);
         }
@@ -143,11 +154,17 @@ public:
             SetEmptyDepthBuffer();
             return {};
         }
+        const auto host_ptr{system.GPU().MemoryManager().GetPointer(gpu_addr)};
+        const auto cache_addr{ToCacheAddr(host_ptr)};
+        if (!cache_addr) {
+            SetEmptyDepthBuffer();
+            return {};
+        }
         const auto depth_params{SurfaceParams::CreateForDepthBuffer(
             system, regs.zeta_width, regs.zeta_height, regs.zeta.format,
             regs.zeta.memory_layout.block_width, regs.zeta.memory_layout.block_height,
             regs.zeta.memory_layout.block_depth, regs.zeta.memory_layout.type)};
-        auto surface_view = GetSurface(gpu_addr, depth_params, preserve_contents, true);
+        auto surface_view = GetSurface(gpu_addr, cache_addr, depth_params, preserve_contents, true);
         if (depth_buffer.target)
             depth_buffer.target->MarkAsRenderTarget(false, NO_RT);
         depth_buffer.target = surface_view.first;
@@ -180,8 +197,16 @@ public:
             return {};
         }
 
-        auto surface_view = GetSurface(gpu_addr, SurfaceParams::CreateForFramebuffer(system, index),
-                                       preserve_contents, true);
+        const auto host_ptr{system.GPU().MemoryManager().GetPointer(gpu_addr)};
+        const auto cache_addr{ToCacheAddr(host_ptr)};
+        if (!cache_addr) {
+            SetEmptyColorBuffer(index);
+            return {};
+        }
+
+        auto surface_view =
+            GetSurface(gpu_addr, cache_addr, SurfaceParams::CreateForFramebuffer(system, index),
+                       preserve_contents, true);
         if (render_targets[index].target)
             render_targets[index].target->MarkAsRenderTarget(false, NO_RT);
         render_targets[index].target = surface_view.first;
@@ -230,8 +255,14 @@ public:
         const GPUVAddr src_gpu_addr = src_config.Address();
         const GPUVAddr dst_gpu_addr = dst_config.Address();
         DeduceBestBlit(src_params, dst_params, src_gpu_addr, dst_gpu_addr);
-        std::pair<TSurface, TView> dst_surface = GetSurface(dst_gpu_addr, dst_params, true, false);
-        std::pair<TSurface, TView> src_surface = GetSurface(src_gpu_addr, src_params, true, false);
+        const auto dst_host_ptr{system.GPU().MemoryManager().GetPointer(dst_gpu_addr)};
+        const auto dst_cache_addr{ToCacheAddr(dst_host_ptr)};
+        const auto src_host_ptr{system.GPU().MemoryManager().GetPointer(src_gpu_addr)};
+        const auto src_cache_addr{ToCacheAddr(src_host_ptr)};
+        std::pair<TSurface, TView> dst_surface =
+            GetSurface(dst_gpu_addr, dst_cache_addr, dst_params, true, false);
+        std::pair<TSurface, TView> src_surface =
+            GetSurface(src_gpu_addr, src_cache_addr, src_params, true, false);
         ImageBlit(src_surface.second, dst_surface.second, copy_config);
         dst_surface.first->MarkAsModified(true, Tick());
     }
@@ -347,13 +378,6 @@ protected:
         return new_surface;
     }
 
-    std::pair<TSurface, TView> GetFermiSurface(
-        const Tegra::Engines::Fermi2D::Regs::Surface& config) {
-        SurfaceParams params = SurfaceParams::CreateForFermiCopySurface(config);
-        const GPUVAddr gpu_addr = config.Address();
-        return GetSurface(gpu_addr, params, true, false);
-    }
-
     Core::System& system;
 
 private:
@@ -614,22 +638,9 @@ private:
      *                          left blank.
      * @param is_render         Whether or not the surface is a render target.
      **/
-    std::pair<TSurface, TView> GetSurface(const GPUVAddr gpu_addr, const SurfaceParams& params,
-                                          bool preserve_contents, bool is_render) {
-        const auto host_ptr{system.GPU().MemoryManager().GetPointer(gpu_addr)};
-        const auto cache_addr{ToCacheAddr(host_ptr)};
-
-        // Step 0: guarantee a valid surface
-        if (!cache_addr) {
-            // Return a null surface if it's invalid
-            SurfaceParams new_params = params;
-            new_params.width = 1;
-            new_params.height = 1;
-            new_params.depth = 1;
-            new_params.block_height = 0;
-            new_params.block_depth = 0;
-            return InitializeSurface(gpu_addr, new_params, false);
-        }
+    std::pair<TSurface, TView> GetSurface(const GPUVAddr gpu_addr, const CacheAddr cache_addr,
+                                          const SurfaceParams& params, bool preserve_contents,
+                                          bool is_render) {
 
         // Step 1
         // Check Level 1 Cache for a fast structural match. If candidate surface
@@ -793,6 +804,41 @@ private:
         }
     }
 
+    /**
+     * Gets a null surface based on a target texture.
+     * @param target The target of the null surface.
+     */
+    TView GetNullSurface(SurfaceTarget target) {
+        const u32 i_target = static_cast<u32>(target);
+        if (const auto it = invalid_cache.find(i_target); it != invalid_cache.end()) {
+            return it->second->GetMainView();
+        }
+        SurfaceParams params{};
+        params.target = target;
+        params.is_tiled = false;
+        params.srgb_conversion = false;
+        params.is_layered = false;
+        params.block_width = 0;
+        params.block_height = 0;
+        params.block_depth = 0;
+        params.tile_width_spacing = 1;
+        params.width = 1;
+        params.height = 1;
+        params.depth = 1;
+        params.pitch = 4;
+        params.num_levels = 1;
+        params.emulated_levels = 1;
+        params.pixel_format = VideoCore::Surface::PixelFormat::RGBA16F;
+        params.type = VideoCore::Surface::SurfaceType::ColorTexture;
+        auto surface = CreateSurface(0ULL, params);
+        invalid_memory.clear();
+        invalid_memory.resize(surface->GetHostSizeInBytes(), 0U);
+        surface->UploadTexture(invalid_memory);
+        surface->MarkAsModified(false, Tick());
+        invalid_cache.emplace(i_target, surface);
+        return surface->GetMainView();
+    }
+
     /**
      * Gets the a source and destination starting address and parameters,
      * and tries to deduce if they are supposed to be depth textures. If so, their
@@ -991,6 +1037,11 @@ private:
 
     std::vector<TSurface> sampled_textures;
 
+    /// This cache stores null surfaces in order to be used as a placeholder
+    /// for invalid texture calls.
+    std::unordered_map<u32, TSurface> invalid_cache;
+    std::vector<u8> invalid_memory;
+
     StagingCache staging_cache;
     std::recursive_mutex mutex;
 };