6 files changed, 70 insertions, 10 deletions

diff --git a/src/common/common_funcs.h b/src/common/common_funcs.h
index 1e74d6930..4c1e29de6 100644
--- a/src/common/common_funcs.h
+++ b/src/common/common_funcs.h
@@ -61,6 +61,14 @@ __declspec(dllimport) void __stdcall DebugBreak(void);
         using T = std::underlying_type_t<type>;                                                    \
         return static_cast<type>(static_cast<T>(a) ^ static_cast<T>(b));                           \
     }                                                                                              \
+    [[nodiscard]] constexpr type operator<<(type a, type b) noexcept {                             \
+        using T = std::underlying_type_t<type>;                                                    \
+        return static_cast<type>(static_cast<T>(a) << static_cast<T>(b));                          \
+    }                                                                                              \
+    [[nodiscard]] constexpr type operator>>(type a, type b) noexcept {                             \
+        using T = std::underlying_type_t<type>;                                                    \
+        return static_cast<type>(static_cast<T>(a) >> static_cast<T>(b));                          \
+    }                                                                                              \
     constexpr type& operator|=(type& a, type b) noexcept {                                         \
         a = a | b;                                                                                 \
         return a;                                                                                  \
@@ -73,6 +81,14 @@ __declspec(dllimport) void __stdcall DebugBreak(void);
         a = a ^ b;                                                                                 \
         return a;                                                                                  \
     }                                                                                              \
+    constexpr type& operator<<=(type& a, type b) noexcept {                                        \
+        a = a << b;                                                                                \
+        return a;                                                                                  \
+    }                                                                                              \
+    constexpr type& operator>>=(type& a, type b) noexcept {                                        \
+        a = a >> b;                                                                                \
+        return a;                                                                                  \
+    }                                                                                              \
     [[nodiscard]] constexpr type operator~(type key) noexcept {                                    \
         using T = std::underlying_type_t<type>;                                                    \
         return static_cast<type>(~static_cast<T>(key));                                            \
diff --git a/src/core/hle/service/filesystem/filesystem.cpp b/src/core/hle/service/filesystem/filesystem.cpp
index db17d61e4..c8d65f328 100644
--- a/src/core/hle/service/filesystem/filesystem.cpp
+++ b/src/core/hle/service/filesystem/filesystem.cpp
@@ -97,6 +97,11 @@ ResultCode VfsDirectoryServiceWrapper::DeleteFile(const std::string& path_) cons
 
 ResultCode VfsDirectoryServiceWrapper::CreateDirectory(const std::string& path_) const {
     std::string path(Common::FS::SanitizePath(path_));
+
+    // NOTE: This is inaccurate behavior. CreateDirectory is not recursive.
+    // CreateDirectory should return PathNotFound if the parent directory does not exist.
+    // This is here temporarily in order to have UMM "work" in the meantime.
+    // TODO (Morph): Remove this when a hardware test verifies the correct behavior.
     const auto components = Common::FS::SplitPathComponents(path);
     std::string relative_path;
     for (const auto& component : components) {
diff --git a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp
index adb557f60..d87da2a34 100644
--- a/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp
+++ b/src/video_core/renderer_vulkan/vk_descriptor_pool.cpp
@@ -19,7 +19,6 @@ namespace Vulkan {
 // Prefer small grow rates to avoid saturating the descriptor pool with barely used pipelines
 constexpr size_t SETS_GROW_RATE = 16;
 constexpr s32 SCORE_THRESHOLD = 3;
-constexpr u32 SETS_PER_POOL = 64;
 
 struct DescriptorBank {
     DescriptorBankInfo info;
@@ -59,11 +58,12 @@ static DescriptorBankInfo MakeBankInfo(std::span<const Shader::Info> infos) {
 static void AllocatePool(const Device& device, DescriptorBank& bank) {
     std::array<VkDescriptorPoolSize, 6> pool_sizes;
     size_t pool_cursor{};
+    const u32 sets_per_pool = device.GetSetsPerPool();
     const auto add = [&](VkDescriptorType type, u32 count) {
         if (count > 0) {
             pool_sizes[pool_cursor++] = {
                 .type = type,
-                .descriptorCount = count * SETS_PER_POOL,
+                .descriptorCount = count * sets_per_pool,
             };
         }
     };
@@ -78,7 +78,7 @@ static void AllocatePool(const Device& device, DescriptorBank& bank) {
         .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
         .pNext = nullptr,
         .flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT,
-        .maxSets = SETS_PER_POOL,
+        .maxSets = sets_per_pool,
         .poolSizeCount = static_cast<u32>(pool_cursor),
         .pPoolSizes = std::data(pool_sizes),
     }));
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index 841a6b846..3bcd6d6cc 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -765,12 +765,7 @@ void RasterizerVulkan::UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs) {
     const Maxwell::StencilOp zpass = regs.stencil_front_op_zpass;
     const Maxwell::ComparisonOp compare = regs.stencil_front_func_func;
     if (regs.stencil_two_side_enable) {
-        scheduler.Record([fail, zfail, zpass, compare](vk::CommandBuffer cmdbuf) {
-            cmdbuf.SetStencilOpEXT(VK_STENCIL_FACE_FRONT_AND_BACK, MaxwellToVK::StencilOp(fail),
-                                   MaxwellToVK::StencilOp(zpass), MaxwellToVK::StencilOp(zfail),
-                                   MaxwellToVK::ComparisonOp(compare));
-        });
-    } else {
+        // Separate stencil op per face
         const Maxwell::StencilOp back_fail = regs.stencil_back_op_fail;
         const Maxwell::StencilOp back_zfail = regs.stencil_back_op_zfail;
         const Maxwell::StencilOp back_zpass = regs.stencil_back_op_zpass;
@@ -785,6 +780,13 @@ void RasterizerVulkan::UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs) {
                                    MaxwellToVK::StencilOp(back_zfail),
                                    MaxwellToVK::ComparisonOp(back_compare));
         });
+    } else {
+        // Front face defines the stencil op of both faces
+        scheduler.Record([fail, zfail, zpass, compare](vk::CommandBuffer cmdbuf) {
+            cmdbuf.SetStencilOpEXT(VK_STENCIL_FACE_FRONT_AND_BACK, MaxwellToVK::StencilOp(fail),
+                                   MaxwellToVK::StencilOp(zpass), MaxwellToVK::StencilOp(zfail),
+                                   MaxwellToVK::ComparisonOp(compare));
+        });
     }
 }
 
diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp
index 24821c1a3..24fb50db9 100644
--- a/src/video_core/vulkan_common/vulkan_device.cpp
+++ b/src/video_core/vulkan_common/vulkan_device.cpp
@@ -243,6 +243,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
     SetupFamilies(surface);
     SetupFeatures();
     SetupProperties();
+    CollectTelemetryParameters();
 
     const auto queue_cis = GetDeviceQueueCreateInfos();
     const std::vector extensions = LoadExtensions(surface != nullptr);
@@ -368,6 +369,18 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
     };
     SetNext(next, demote);
 
+    if (driver_id == VK_DRIVER_ID_AMD_PROPRIETARY || driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE) {
+        const u32 version = properties.driverVersion;
+        // Broken in this driver
+        if (version > VK_MAKE_API_VERSION(0, 2, 0, 193)) {
+            LOG_WARNING(Render_Vulkan, "AMD proprietary driver versions newer than 21.9.1 "
+                                       "(windows) / 0.2.0.194 (amdvlk) have "
+                                       "broken VkPhysicalDeviceFloat16Int8FeaturesKHR");
+            is_int8_supported = false;
+            is_float16_supported = false;
+        }
+    }
+
     if (is_int8_supported || is_float16_supported) {
         VkPhysicalDeviceFloat16Int8FeaturesKHR float16_int8{
             .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR,
@@ -560,7 +573,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
     logical = vk::Device::Create(physical, queue_cis, extensions, first_next, dld);
 
     CollectPhysicalMemoryInfo();
-    CollectTelemetryParameters();
     CollectToolingInfo();
 
     if (driver_id == VK_DRIVER_ID_NVIDIA_PROPRIETARY_KHR) {
@@ -587,6 +599,26 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
             ext_extended_dynamic_state = false;
         }
     }
+
+    sets_per_pool = 64;
+    if (driver_id == VK_DRIVER_ID_AMD_PROPRIETARY || driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE) {
+        // AMD drivers need a higher amount of Sets per Pool in certain circunstances like in XC2.
+        sets_per_pool = 96;
+    }
+
+    const bool is_amd = driver_id == VK_DRIVER_ID_AMD_PROPRIETARY ||
+                        driver_id == VK_DRIVER_ID_MESA_RADV ||
+                        driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE;
+    if (ext_sampler_filter_minmax && is_amd) {
+        // Disable ext_sampler_filter_minmax on AMD GCN4 and lower as it is broken.
+        if (!is_float16_supported) {
+            LOG_WARNING(
+                Render_Vulkan,
+                "Blacklisting AMD GCN4 and lower for VK_EXT_SAMPLER_FILTER_MINMAX_EXTENSION_NAME");
+            ext_sampler_filter_minmax = false;
+        }
+    }
+
     if (ext_vertex_input_dynamic_state && driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS) {
         LOG_WARNING(Render_Vulkan, "Blacklisting Intel for VK_EXT_vertex_input_dynamic_state");
         ext_vertex_input_dynamic_state = false;
diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h
index 5599c38c5..bc180a32a 100644
--- a/src/video_core/vulkan_common/vulkan_device.h
+++ b/src/video_core/vulkan_common/vulkan_device.h
@@ -323,6 +323,10 @@ public:
         return device_access_memory;
     }
 
+    u32 GetSetsPerPool() const {
+        return sets_per_pool;
+    }
+
 private:
     /// Checks if the physical device is suitable.
     void CheckSuitability(bool requires_swapchain) const;
@@ -376,6 +380,7 @@ private:
     VkShaderStageFlags guest_warp_stages{};     ///< Stages where the guest warp size can be forced.
     u64 device_access_memory{};                 ///< Total size of device local memory in bytes.
     u32 max_push_descriptors{};                 ///< Maximum number of push descriptors
+    u32 sets_per_pool{};                        ///< Sets per Description Pool
     bool is_optimal_astc_supported{};           ///< Support for native ASTC.
     bool is_float16_supported{};                ///< Support for float16 arithmetic.
     bool is_int8_supported{};                   ///< Support for int8 arithmetic.