Merge branch 'master' into ssbo-align

18 files changed, 282 insertions, 130 deletions

diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
index 7e7a80740..c4c30d807 100644
--- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp
+++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
@@ -132,16 +132,12 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
     const bool use_accelerated =
         rasterizer.AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride);
     const bool is_srgb = use_accelerated && screen_info.is_srgb;
+    RenderScreenshot(*framebuffer, use_accelerated);
 
-    {
-        std::scoped_lock lock{rasterizer.LockCaches()};
-        RenderScreenshot(*framebuffer, use_accelerated);
-
-        Frame* frame = present_manager.GetRenderFrame();
-        blit_screen.DrawToSwapchain(frame, *framebuffer, use_accelerated, is_srgb);
-        scheduler.Flush(*frame->render_ready);
-        present_manager.Present(frame);
-    }
+    Frame* frame = present_manager.GetRenderFrame();
+    blit_screen.DrawToSwapchain(frame, *framebuffer, use_accelerated, is_srgb);
+    scheduler.Flush(*frame->render_ready);
+    present_manager.Present(frame);
 
     gpu.RendererFrameEndNotify();
     rasterizer.TickFrame();
diff --git a/src/video_core/renderer_vulkan/vk_blit_screen.cpp b/src/video_core/renderer_vulkan/vk_blit_screen.cpp
index 52fc142d1..66483a900 100644
--- a/src/video_core/renderer_vulkan/vk_blit_screen.cpp
+++ b/src/video_core/renderer_vulkan/vk_blit_screen.cpp
@@ -137,6 +137,56 @@ BlitScreen::BlitScreen(Core::Memory::Memory& cpu_memory_, Core::Frontend::EmuWin
 
 BlitScreen::~BlitScreen() = default;
 
+static Common::Rectangle<f32> NormalizeCrop(const Tegra::FramebufferConfig& framebuffer,
+                                            const ScreenInfo& screen_info) {
+    f32 left, top, right, bottom;
+
+    if (!framebuffer.crop_rect.IsEmpty()) {
+        // If crop rectangle is not empty, apply properties from rectangle.
+        left = static_cast<f32>(framebuffer.crop_rect.left);
+        top = static_cast<f32>(framebuffer.crop_rect.top);
+        right = static_cast<f32>(framebuffer.crop_rect.right);
+        bottom = static_cast<f32>(framebuffer.crop_rect.bottom);
+    } else {
+        // Otherwise, fall back to framebuffer dimensions.
+        left = 0;
+        top = 0;
+        right = static_cast<f32>(framebuffer.width);
+        bottom = static_cast<f32>(framebuffer.height);
+    }
+
+    // Apply transformation flags.
+    auto framebuffer_transform_flags = framebuffer.transform_flags;
+
+    if (True(framebuffer_transform_flags & Service::android::BufferTransformFlags::FlipH)) {
+        // Switch left and right.
+        std::swap(left, right);
+    }
+    if (True(framebuffer_transform_flags & Service::android::BufferTransformFlags::FlipV)) {
+        // Switch top and bottom.
+        std::swap(top, bottom);
+    }
+
+    framebuffer_transform_flags &= ~Service::android::BufferTransformFlags::FlipH;
+    framebuffer_transform_flags &= ~Service::android::BufferTransformFlags::FlipV;
+    if (True(framebuffer_transform_flags)) {
+        UNIMPLEMENTED_MSG("Unsupported framebuffer_transform_flags={}",
+                          static_cast<u32>(framebuffer_transform_flags));
+    }
+
+    // Get the screen properties.
+    const f32 screen_width = static_cast<f32>(screen_info.width);
+    const f32 screen_height = static_cast<f32>(screen_info.height);
+
+    // Normalize coordinate space.
+    left /= screen_width;
+    top /= screen_height;
+    right /= screen_width;
+    bottom /= screen_height;
+
+    return Common::Rectangle<f32>(left, top, right, bottom);
+}
+
 void BlitScreen::Recreate() {
     present_manager.WaitPresent();
     scheduler.Finish();
@@ -354,17 +404,10 @@ void BlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
         source_image_view = smaa->Draw(scheduler, image_index, source_image, source_image_view);
     }
     if (fsr) {
-        auto crop_rect = framebuffer.crop_rect;
-        if (crop_rect.GetWidth() == 0) {
-            crop_rect.right = framebuffer.width;
-        }
-        if (crop_rect.GetHeight() == 0) {
-            crop_rect.bottom = framebuffer.height;
-        }
-        crop_rect = crop_rect.Scale(Settings::values.resolution_info.up_factor);
-        VkExtent2D fsr_input_size{
-            .width = Settings::values.resolution_info.ScaleUp(framebuffer.width),
-            .height = Settings::values.resolution_info.ScaleUp(framebuffer.height),
+        const auto crop_rect = NormalizeCrop(framebuffer, screen_info);
+        const VkExtent2D fsr_input_size{
+            .width = Settings::values.resolution_info.ScaleUp(screen_info.width),
+            .height = Settings::values.resolution_info.ScaleUp(screen_info.height),
         };
         VkImageView fsr_image_view =
             fsr->Draw(scheduler, image_index, source_image_view, fsr_input_size, crop_rect);
@@ -1397,61 +1440,37 @@ void BlitScreen::SetUniformData(BufferData& data, const Layout::FramebufferLayou
 
 void BlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer,
                                const Layout::FramebufferLayout layout) const {
-    const auto& framebuffer_transform_flags = framebuffer.transform_flags;
-    const auto& framebuffer_crop_rect = framebuffer.crop_rect;
-
-    static constexpr Common::Rectangle<f32> texcoords{0.f, 0.f, 1.f, 1.f};
-    auto left = texcoords.left;
-    auto right = texcoords.right;
-
-    switch (framebuffer_transform_flags) {
-    case Service::android::BufferTransformFlags::Unset:
-        break;
-    case Service::android::BufferTransformFlags::FlipV:
-        // Flip the framebuffer vertically
-        left = texcoords.right;
-        right = texcoords.left;
-        break;
-    default:
-        UNIMPLEMENTED_MSG("Unsupported framebuffer_transform_flags={}",
-                          static_cast<u32>(framebuffer_transform_flags));
-        break;
-    }
+    f32 left, top, right, bottom;
 
-    UNIMPLEMENTED_IF(framebuffer_crop_rect.left != 0);
-
-    f32 left_start{};
-    if (framebuffer_crop_rect.Top() > 0) {
-        left_start = static_cast<f32>(framebuffer_crop_rect.Top()) /
-                     static_cast<f32>(framebuffer_crop_rect.Bottom());
-    }
-    f32 scale_u = static_cast<f32>(framebuffer.width) / static_cast<f32>(screen_info.width);
-    f32 scale_v = static_cast<f32>(framebuffer.height) / static_cast<f32>(screen_info.height);
-    // Scale the output by the crop width/height. This is commonly used with 1280x720 rendering
-    // (e.g. handheld mode) on a 1920x1080 framebuffer.
-    if (!fsr) {
-        if (framebuffer_crop_rect.GetWidth() > 0) {
-            scale_u = static_cast<f32>(framebuffer_crop_rect.GetWidth()) /
-                      static_cast<f32>(screen_info.width);
-        }
-        if (framebuffer_crop_rect.GetHeight() > 0) {
-            scale_v = static_cast<f32>(framebuffer_crop_rect.GetHeight()) /
-                      static_cast<f32>(screen_info.height);
-        }
+    if (fsr) {
+        // FSR has already applied the crop, so we just want to render the image
+        // it has produced.
+        left = 0;
+        top = 0;
+        right = 1;
+        bottom = 1;
+    } else {
+        // Get the normalized crop rectangle.
+        const auto crop = NormalizeCrop(framebuffer, screen_info);
+
+        // Apply the crop.
+        left = crop.left;
+        top = crop.top;
+        right = crop.right;
+        bottom = crop.bottom;
     }
 
+    // Map the coordinates to the screen.
     const auto& screen = layout.screen;
     const auto x = static_cast<f32>(screen.left);
     const auto y = static_cast<f32>(screen.top);
     const auto w = static_cast<f32>(screen.GetWidth());
     const auto h = static_cast<f32>(screen.GetHeight());
-    data.vertices[0] = ScreenRectVertex(x, y, texcoords.top * scale_u, left_start + left * scale_v);
-    data.vertices[1] =
-        ScreenRectVertex(x + w, y, texcoords.bottom * scale_u, left_start + left * scale_v);
-    data.vertices[2] =
-        ScreenRectVertex(x, y + h, texcoords.top * scale_u, left_start + right * scale_v);
-    data.vertices[3] =
-        ScreenRectVertex(x + w, y + h, texcoords.bottom * scale_u, left_start + right * scale_v);
+
+    data.vertices[0] = ScreenRectVertex(x, y, left, top);
+    data.vertices[1] = ScreenRectVertex(x + w, y, right, top);
+    data.vertices[2] = ScreenRectVertex(x, y + h, left, bottom);
+    data.vertices[3] = ScreenRectVertex(x + w, y + h, right, bottom);
 }
 
 void BlitScreen::CreateSMAA(VkExtent2D smaa_size) {
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
index 976c3f6a6..5958f52f7 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
@@ -79,13 +79,13 @@ vk::Buffer CreateBuffer(const Device& device, const MemoryAllocator& memory_allo
 } // Anonymous namespace
 
 Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params)
-    : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params) {}
+    : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params), tracker{4096} {}
 
 Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
                VAddr cpu_addr_, u64 size_bytes_)
     : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(rasterizer_, cpu_addr_, size_bytes_),
-      device{&runtime.device}, buffer{
-                                   CreateBuffer(*device, runtime.memory_allocator, SizeBytes())} {
+      device{&runtime.device}, buffer{CreateBuffer(*device, runtime.memory_allocator, SizeBytes())},
+      tracker{SizeBytes()} {
     if (runtime.device.HasDebuggingToolAttached()) {
         buffer.SetObjectNameEXT(fmt::format("Buffer 0x{:x}", CpuAddr()).c_str());
     }
@@ -359,12 +359,31 @@ u32 BufferCacheRuntime::GetStorageBufferAlignment() const {
     return static_cast<u32>(device.GetStorageBufferAlignment());
 }
 
+void BufferCacheRuntime::TickFrame(VideoCommon::SlotVector<Buffer>& slot_buffers) noexcept {
+    for (auto it = slot_buffers.begin(); it != slot_buffers.end(); it++) {
+        it->ResetUsageTracking();
+    }
+}
+
 void BufferCacheRuntime::Finish() {
     scheduler.Finish();
 }
 
+bool BufferCacheRuntime::CanReorderUpload(const Buffer& buffer,
+                                          std::span<const VideoCommon::BufferCopy> copies) {
+    if (Settings::values.disable_buffer_reorder) {
+        return false;
+    }
+    const bool can_use_upload_cmdbuf =
+        std::ranges::all_of(copies, [&](const VideoCommon::BufferCopy& copy) {
+            return !buffer.IsRegionUsed(copy.dst_offset, copy.size);
+        });
+    return can_use_upload_cmdbuf;
+}
+
 void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer,
-                                    std::span<const VideoCommon::BufferCopy> copies, bool barrier) {
+                                    std::span<const VideoCommon::BufferCopy> copies, bool barrier,
+                                    bool can_reorder_upload) {
     if (dst_buffer == VK_NULL_HANDLE || src_buffer == VK_NULL_HANDLE) {
         return;
     }
@@ -380,9 +399,18 @@ void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer,
         .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
         .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
     };
+
     // Measuring a popular game, this number never exceeds the specified size once data is warmed up
     boost::container::small_vector<VkBufferCopy, 8> vk_copies(copies.size());
     std::ranges::transform(copies, vk_copies.begin(), MakeBufferCopy);
+    if (src_buffer == staging_pool.StreamBuf() && can_reorder_upload) {
+        scheduler.RecordWithUploadBuffer([src_buffer, dst_buffer, vk_copies](
+                                             vk::CommandBuffer, vk::CommandBuffer upload_cmdbuf) {
+            upload_cmdbuf.CopyBuffer(src_buffer, dst_buffer, vk_copies);
+        });
+        return;
+    }
+
     scheduler.RequestOutsideRenderPassOperationContext();
     scheduler.Record([src_buffer, dst_buffer, vk_copies, barrier](vk::CommandBuffer cmdbuf) {
         if (barrier) {
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.h b/src/video_core/renderer_vulkan/vk_buffer_cache.h
index 833dfac45..0b3fbd6d0 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.h
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.h
@@ -5,6 +5,7 @@
 
 #include "video_core/buffer_cache/buffer_cache_base.h"
 #include "video_core/buffer_cache/memory_tracker_base.h"
+#include "video_core/buffer_cache/usage_tracker.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/vk_compute_pass.h"
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
@@ -34,6 +35,18 @@ public:
         return *buffer;
     }
 
+    [[nodiscard]] bool IsRegionUsed(u64 offset, u64 size) const noexcept {
+        return tracker.IsUsed(offset, size);
+    }
+
+    void MarkUsage(u64 offset, u64 size) noexcept {
+        tracker.Track(offset, size);
+    }
+
+    void ResetUsageTracking() noexcept {
+        tracker.Reset();
+    }
+
     operator VkBuffer() const noexcept {
         return *buffer;
     }
@@ -49,6 +62,7 @@ private:
     const Device* device{};
     vk::Buffer buffer;
     std::vector<BufferView> views;
+    VideoCommon::UsageTracker tracker;
 };
 
 class QuadArrayIndexBuffer;
@@ -67,6 +81,8 @@ public:
                                 ComputePassDescriptorQueue& compute_pass_descriptor_queue,
                                 DescriptorPool& descriptor_pool);
 
+    void TickFrame(VideoCommon::SlotVector<Buffer>& slot_buffers) noexcept;
+
     void Finish();
 
     u64 GetDeviceLocalMemory() const;
@@ -81,12 +97,15 @@ public:
 
     [[nodiscard]] StagingBufferRef DownloadStagingBuffer(size_t size, bool deferred = false);
 
+    bool CanReorderUpload(const Buffer& buffer, std::span<const VideoCommon::BufferCopy> copies);
+
     void FreeDeferredStagingBuffer(StagingBufferRef& ref);
 
     void PreCopyBarrier();
 
     void CopyBuffer(VkBuffer src_buffer, VkBuffer dst_buffer,
-                    std::span<const VideoCommon::BufferCopy> copies, bool barrier = true);
+                    std::span<const VideoCommon::BufferCopy> copies, bool barrier,
+                    bool can_reorder_upload = false);
 
     void PostCopyBarrier();
 
diff --git a/src/video_core/renderer_vulkan/vk_fsr.cpp b/src/video_core/renderer_vulkan/vk_fsr.cpp
index ce8f3f3c2..f7a05fbc0 100644
--- a/src/video_core/renderer_vulkan/vk_fsr.cpp
+++ b/src/video_core/renderer_vulkan/vk_fsr.cpp
@@ -34,7 +34,7 @@ FSR::FSR(const Device& device_, MemoryAllocator& memory_allocator_, size_t image
 }
 
 VkImageView FSR::Draw(Scheduler& scheduler, size_t image_index, VkImageView image_view,
-                      VkExtent2D input_image_extent, const Common::Rectangle<int>& crop_rect) {
+                      VkExtent2D input_image_extent, const Common::Rectangle<f32>& crop_rect) {
 
     UpdateDescriptorSet(image_index, image_view);
 
@@ -61,15 +61,21 @@ VkImageView FSR::Draw(Scheduler& scheduler, size_t image_index, VkImageView imag
 
         cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *easu_pipeline);
 
+        const f32 input_image_width = static_cast<f32>(input_image_extent.width);
+        const f32 input_image_height = static_cast<f32>(input_image_extent.height);
+        const f32 output_image_width = static_cast<f32>(output_size.width);
+        const f32 output_image_height = static_cast<f32>(output_size.height);
+        const f32 viewport_width = (crop_rect.right - crop_rect.left) * input_image_width;
+        const f32 viewport_x = crop_rect.left * input_image_width;
+        const f32 viewport_height = (crop_rect.bottom - crop_rect.top) * input_image_height;
+        const f32 viewport_y = crop_rect.top * input_image_height;
+
         std::array<u32, 4 * 4> push_constants;
-        FsrEasuConOffset(
-            push_constants.data() + 0, push_constants.data() + 4, push_constants.data() + 8,
-            push_constants.data() + 12,
-
-            static_cast<f32>(crop_rect.GetWidth()), static_cast<f32>(crop_rect.GetHeight()),
-            static_cast<f32>(input_image_extent.width), static_cast<f32>(input_image_extent.height),
-            static_cast<f32>(output_size.width), static_cast<f32>(output_size.height),
-            static_cast<f32>(crop_rect.left), static_cast<f32>(crop_rect.top));
+        FsrEasuConOffset(push_constants.data() + 0, push_constants.data() + 4,
+                         push_constants.data() + 8, push_constants.data() + 12,
+
+                         viewport_width, viewport_height, input_image_width, input_image_height,
+                         output_image_width, output_image_height, viewport_x, viewport_y);
         cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, push_constants);
 
         {
diff --git a/src/video_core/renderer_vulkan/vk_fsr.h b/src/video_core/renderer_vulkan/vk_fsr.h
index 8bb9fc23a..3505c1416 100644
--- a/src/video_core/renderer_vulkan/vk_fsr.h
+++ b/src/video_core/renderer_vulkan/vk_fsr.h
@@ -17,7 +17,7 @@ public:
     explicit FSR(const Device& device, MemoryAllocator& memory_allocator, size_t image_count,
                  VkExtent2D output_size);
     VkImageView Draw(Scheduler& scheduler, size_t image_index, VkImageView image_view,
-                     VkExtent2D input_image_extent, const Common::Rectangle<int>& crop_rect);
+                     VkExtent2D input_image_extent, const Common::Rectangle<f32>& crop_rect);
 
 private:
     void CreateDescriptorPool();
diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.cpp b/src/video_core/renderer_vulkan/vk_master_semaphore.cpp
index 6b288b994..ac8b6e838 100644
--- a/src/video_core/renderer_vulkan/vk_master_semaphore.cpp
+++ b/src/video_core/renderer_vulkan/vk_master_semaphore.cpp
@@ -100,12 +100,14 @@ void MasterSemaphore::Wait(u64 tick) {
     Refresh();
 }
 
-VkResult MasterSemaphore::SubmitQueue(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
-                                      VkSemaphore wait_semaphore, u64 host_tick) {
+VkResult MasterSemaphore::SubmitQueue(vk::CommandBuffer& cmdbuf, vk::CommandBuffer& upload_cmdbuf,
+                                      VkSemaphore signal_semaphore, VkSemaphore wait_semaphore,
+                                      u64 host_tick) {
     if (semaphore) {
-        return SubmitQueueTimeline(cmdbuf, signal_semaphore, wait_semaphore, host_tick);
+        return SubmitQueueTimeline(cmdbuf, upload_cmdbuf, signal_semaphore, wait_semaphore,
+                                   host_tick);
     } else {
-        return SubmitQueueFence(cmdbuf, signal_semaphore, wait_semaphore, host_tick);
+        return SubmitQueueFence(cmdbuf, upload_cmdbuf, signal_semaphore, wait_semaphore, host_tick);
     }
 }
 
@@ -115,6 +117,7 @@ static constexpr std::array<VkPipelineStageFlags, 2> wait_stage_masks{
 };
 
 VkResult MasterSemaphore::SubmitQueueTimeline(vk::CommandBuffer& cmdbuf,
+                                              vk::CommandBuffer& upload_cmdbuf,
                                               VkSemaphore signal_semaphore,
                                               VkSemaphore wait_semaphore, u64 host_tick) {
     const VkSemaphore timeline_semaphore = *semaphore;
@@ -123,6 +126,8 @@ VkResult MasterSemaphore::SubmitQueueTimeline(vk::CommandBuffer& cmdbuf,
     const std::array signal_values{host_tick, u64(0)};
     const std::array signal_semaphores{timeline_semaphore, signal_semaphore};
 
+    const std::array cmdbuffers{*upload_cmdbuf, *cmdbuf};
+
     const u32 num_wait_semaphores = wait_semaphore ? 1 : 0;
     const VkTimelineSemaphoreSubmitInfo timeline_si{
         .sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO,
@@ -138,8 +143,8 @@ VkResult MasterSemaphore::SubmitQueueTimeline(vk::CommandBuffer& cmdbuf,
         .waitSemaphoreCount = num_wait_semaphores,
         .pWaitSemaphores = &wait_semaphore,
         .pWaitDstStageMask = wait_stage_masks.data(),
-        .commandBufferCount = 1,
-        .pCommandBuffers = cmdbuf.address(),
+        .commandBufferCount = static_cast<u32>(cmdbuffers.size()),
+        .pCommandBuffers = cmdbuffers.data(),
         .signalSemaphoreCount = num_signal_semaphores,
         .pSignalSemaphores = signal_semaphores.data(),
     };
@@ -147,19 +152,23 @@ VkResult MasterSemaphore::SubmitQueueTimeline(vk::CommandBuffer& cmdbuf,
     return device.GetGraphicsQueue().Submit(submit_info);
 }
 
-VkResult MasterSemaphore::SubmitQueueFence(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
-                                           VkSemaphore wait_semaphore, u64 host_tick) {
+VkResult MasterSemaphore::SubmitQueueFence(vk::CommandBuffer& cmdbuf,
+                                           vk::CommandBuffer& upload_cmdbuf,
+                                           VkSemaphore signal_semaphore, VkSemaphore wait_semaphore,
+                                           u64 host_tick) {
     const u32 num_signal_semaphores = signal_semaphore ? 1 : 0;
     const u32 num_wait_semaphores = wait_semaphore ? 1 : 0;
 
+    const std::array cmdbuffers{*upload_cmdbuf, *cmdbuf};
+
     const VkSubmitInfo submit_info{
         .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
         .pNext = nullptr,
         .waitSemaphoreCount = num_wait_semaphores,
         .pWaitSemaphores = &wait_semaphore,
         .pWaitDstStageMask = wait_stage_masks.data(),
-        .commandBufferCount = 1,
-        .pCommandBuffers = cmdbuf.address(),
+        .commandBufferCount = static_cast<u32>(cmdbuffers.size()),
+        .pCommandBuffers = cmdbuffers.data(),
         .signalSemaphoreCount = num_signal_semaphores,
         .pSignalSemaphores = &signal_semaphore,
     };
diff --git a/src/video_core/renderer_vulkan/vk_master_semaphore.h b/src/video_core/renderer_vulkan/vk_master_semaphore.h
index 3f599d7bd..7dfb93ffb 100644
--- a/src/video_core/renderer_vulkan/vk_master_semaphore.h
+++ b/src/video_core/renderer_vulkan/vk_master_semaphore.h
@@ -52,14 +52,16 @@ public:
     void Wait(u64 tick);
 
     /// Submits the device graphics queue, updating the tick as necessary
-    VkResult SubmitQueue(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
-                         VkSemaphore wait_semaphore, u64 host_tick);
+    VkResult SubmitQueue(vk::CommandBuffer& cmdbuf, vk::CommandBuffer& upload_cmdbuf,
+                         VkSemaphore signal_semaphore, VkSemaphore wait_semaphore, u64 host_tick);
 
 private:
-    VkResult SubmitQueueTimeline(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
-                                 VkSemaphore wait_semaphore, u64 host_tick);
-    VkResult SubmitQueueFence(vk::CommandBuffer& cmdbuf, VkSemaphore signal_semaphore,
-                              VkSemaphore wait_semaphore, u64 host_tick);
+    VkResult SubmitQueueTimeline(vk::CommandBuffer& cmdbuf, vk::CommandBuffer& upload_cmdbuf,
+                                 VkSemaphore signal_semaphore, VkSemaphore wait_semaphore,
+                                 u64 host_tick);
+    VkResult SubmitQueueFence(vk::CommandBuffer& cmdbuf, vk::CommandBuffer& upload_cmdbuf,
+                              VkSemaphore signal_semaphore, VkSemaphore wait_semaphore,
+                              u64 host_tick);
 
     void WaitThread(std::stop_token token);
 
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index 0d604eee3..2a13b2a72 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -263,6 +263,22 @@ Shader::RuntimeInfo MakeRuntimeInfo(std::span<const Shader::IR::Program> program
     info.y_negate = key.state.y_negate != 0;
     return info;
 }
+
+size_t GetTotalPipelineWorkers() {
+    const size_t max_core_threads =
+        std::max<size_t>(static_cast<size_t>(std::thread::hardware_concurrency()), 2ULL) - 1ULL;
+#ifdef ANDROID
+    // Leave at least a few cores free in android
+    constexpr size_t free_cores = 3ULL;
+    if (max_core_threads <= free_cores) {
+        return 1ULL;
+    }
+    return max_core_threads - free_cores;
+#else
+    return max_core_threads;
+#endif
+}
+
 } // Anonymous namespace
 
 size_t ComputePipelineCacheKey::Hash() const noexcept {
@@ -294,11 +310,8 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device
       texture_cache{texture_cache_}, shader_notify{shader_notify_},
       use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()},
       use_vulkan_pipeline_cache{Settings::values.use_vulkan_driver_pipeline_cache.GetValue()},
-#ifdef ANDROID
-      workers(1, "VkPipelineBuilder"),
-#else
-      workers(std::max(std::thread::hardware_concurrency(), 2U) - 1, "VkPipelineBuilder"),
-#endif
+      workers(device.HasBrokenParallelShaderCompiling() ? 1ULL : GetTotalPipelineWorkers(),
+              "VkPipelineBuilder"),
       serialization_thread(1, "VkPipelineSerialization") {
     const auto& float_control{device.FloatControlProperties()};
     const VkDriverId driver_id{device.GetDriverID()};
@@ -338,6 +351,7 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device
         .support_geometry_shader_passthrough = device.IsNvGeometryShaderPassthroughSupported(),
         .support_native_ndc = device.IsExtDepthClipControlSupported(),
         .support_scaled_attributes = !device.MustEmulateScaledFormats(),
+        .support_multi_viewport = device.SupportsMultiViewport(),
 
         .warp_size_potentially_larger_than_guest = device.IsWarpSizePotentiallyBiggerThanGuest(),
 
diff --git a/src/video_core/renderer_vulkan/vk_query_cache.cpp b/src/video_core/renderer_vulkan/vk_query_cache.cpp
index 66c03bf17..078777cdd 100644
--- a/src/video_core/renderer_vulkan/vk_query_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp
@@ -211,6 +211,13 @@ public:
             return;
         }
         PauseCounter();
+        const auto driver_id = device.GetDriverID();
+        if (driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY ||
+            driver_id == VK_DRIVER_ID_ARM_PROPRIETARY || driver_id == VK_DRIVER_ID_MESA_TURNIP) {
+            pending_sync.clear();
+            sync_values_stash.clear();
+            return;
+        }
         sync_values_stash.clear();
         sync_values_stash.emplace_back();
         std::vector<HostSyncValues>* sync_values = &sync_values_stash.back();
@@ -1378,6 +1385,12 @@ bool QueryCacheRuntime::HostConditionalRenderingCompareValues(VideoCommon::Looku
         return true;
     }
 
+    auto driver_id = impl->device.GetDriverID();
+    if (driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY ||
+        driver_id == VK_DRIVER_ID_ARM_PROPRIETARY || driver_id == VK_DRIVER_ID_MESA_TURNIP) {
+        return true;
+    }
+
     for (size_t i = 0; i < 2; i++) {
         is_null[i] = !is_in_ac[i] && check_value(objects[i]->address);
     }
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index 059b7cb40..e0ab1eaac 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -82,7 +82,7 @@ VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t in
     }
 
     if (y_negate) {
-        y += height;
+        y += conv(static_cast<f32>(regs.surface_clip.height));
         height = -height;
     }
 
@@ -199,7 +199,7 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) {
     if (!pipeline) {
         return;
     }
-    std::scoped_lock lock{LockCaches()};
+    std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
     // update engine as channel may be different.
     pipeline->SetEngine(maxwell3d, gpu_memory);
     pipeline->Configure(is_indexed);
@@ -621,7 +621,7 @@ void RasterizerVulkan::OnCacheInvalidation(VAddr addr, u64 size) {
     }
     {
         std::scoped_lock lock{buffer_cache.mutex};
-        buffer_cache.CachedWriteMemory(addr, size);
+        buffer_cache.WriteMemory(addr, size);
     }
     pipeline_cache.InvalidateRegion(addr, size);
 }
@@ -710,7 +710,6 @@ void RasterizerVulkan::TiledCacheBarrier() {
 }
 
 void RasterizerVulkan::FlushCommands() {
-    std::scoped_lock lock{LockCaches()};
     if (draw_counter == 0) {
         return;
     }
@@ -808,7 +807,6 @@ void RasterizerVulkan::FlushWork() {
     if ((++draw_counter & 7) != 7) {
         return;
     }
-    std::scoped_lock lock{LockCaches()};
     if (draw_counter < DRAWS_TO_DISPATCH) {
         // Send recorded tasks to the worker thread
         scheduler.DispatchWork();
@@ -923,9 +921,13 @@ void RasterizerVulkan::UpdateDynamicStates() {
 }
 
 void RasterizerVulkan::HandleTransformFeedback() {
+    static std::once_flag warn_unsupported;
+
     const auto& regs = maxwell3d->regs;
     if (!device.IsExtTransformFeedbackSupported()) {
-        LOG_ERROR(Render_Vulkan, "Transform feedbacks used but not supported");
+        std::call_once(warn_unsupported, [&] {
+            LOG_ERROR(Render_Vulkan, "Transform feedbacks used but not supported");
+        });
         return;
     }
     query_cache.CounterEnable(VideoCommon::QueryType::StreamingByteCount,
@@ -1503,7 +1505,7 @@ void RasterizerVulkan::UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs)
 void RasterizerVulkan::InitializeChannel(Tegra::Control::ChannelState& channel) {
     CreateChannel(channel);
     {
-        std::scoped_lock lock{LockCaches()};
+        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
         texture_cache.CreateChannel(channel);
         buffer_cache.CreateChannel(channel);
     }
@@ -1516,7 +1518,7 @@ void RasterizerVulkan::BindChannel(Tegra::Control::ChannelState& channel) {
     const s32 channel_id = channel.bind_id;
     BindToChannel(channel_id);
     {
-        std::scoped_lock lock{LockCaches()};
+        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
         texture_cache.BindToChannel(channel_id);
         buffer_cache.BindToChannel(channel_id);
     }
@@ -1529,7 +1531,7 @@ void RasterizerVulkan::BindChannel(Tegra::Control::ChannelState& channel) {
 void RasterizerVulkan::ReleaseChannel(s32 channel_id) {
     EraseChannel(channel_id);
     {
-        std::scoped_lock lock{LockCaches()};
+        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
         texture_cache.EraseChannel(channel_id);
         buffer_cache.EraseChannel(channel_id);
     }
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h
index ce3dfbaab..ad069556c 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.h
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.h
@@ -133,10 +133,6 @@ public:
 
     void ReleaseChannel(s32 channel_id) override;
 
-    std::scoped_lock<std::recursive_mutex, std::recursive_mutex> LockCaches() {
-        return std::scoped_lock{buffer_cache.mutex, texture_cache.mutex};
-    }
-
 private:
     static constexpr size_t MAX_TEXTURES = 192;
     static constexpr size_t MAX_IMAGES = 48;
diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp
index 3be7837f4..146923db4 100644
--- a/src/video_core/renderer_vulkan/vk_scheduler.cpp
+++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp
@@ -22,11 +22,12 @@ namespace Vulkan {
 
 MICROPROFILE_DECLARE(Vulkan_WaitForWorker);
 
-void Scheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf) {
+void Scheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf,
+                                         vk::CommandBuffer upload_cmdbuf) {
     auto command = first;
     while (command != nullptr) {
         auto next = command->GetNext();
-        command->Execute(cmdbuf);
+        command->Execute(cmdbuf, upload_cmdbuf);
         command->~Command();
         command = next;
     }
@@ -180,7 +181,7 @@ void Scheduler::WorkerThread(std::stop_token stop_token) {
             // Perform the work, tracking whether the chunk was a submission
             // before executing.
             const bool has_submit = work->HasSubmit();
-            work->ExecuteAll(current_cmdbuf);
+            work->ExecuteAll(current_cmdbuf, current_upload_cmdbuf);
 
             // If the chunk was a submission, reallocate the command buffer.
             if (has_submit) {
@@ -205,6 +206,13 @@ void Scheduler::AllocateWorkerCommandBuffer() {
         .flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
         .pInheritanceInfo = nullptr,
     });
+    current_upload_cmdbuf = vk::CommandBuffer(command_pool->Commit(), device.GetDispatchLoader());
+    current_upload_cmdbuf.Begin({
+        .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO,
+        .pNext = nullptr,
+        .flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
+        .pInheritanceInfo = nullptr,
+    });
 }
 
 u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_semaphore) {
@@ -212,7 +220,17 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se
     InvalidateState();
 
     const u64 signal_value = master_semaphore->NextTick();
-    Record([signal_semaphore, wait_semaphore, signal_value, this](vk::CommandBuffer cmdbuf) {
+    RecordWithUploadBuffer([signal_semaphore, wait_semaphore, signal_value,
+                            this](vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf) {
+        static constexpr VkMemoryBarrier WRITE_BARRIER{
+            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
+        };
+        upload_cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
+                                      VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, WRITE_BARRIER);
+        upload_cmdbuf.End();
         cmdbuf.End();
 
         if (on_submit) {
@@ -221,7 +239,7 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se
 
         std::scoped_lock lock{submit_mutex};
         switch (const VkResult result = master_semaphore->SubmitQueue(
-                    cmdbuf, signal_semaphore, wait_semaphore, signal_value)) {
+                    cmdbuf, upload_cmdbuf, signal_semaphore, wait_semaphore, signal_value)) {
         case VK_SUCCESS:
             break;
         case VK_ERROR_DEVICE_LOST:
diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h
index da03803aa..f8d8ca80a 100644
--- a/src/video_core/renderer_vulkan/vk_scheduler.h
+++ b/src/video_core/renderer_vulkan/vk_scheduler.h
@@ -80,7 +80,8 @@ public:
 
     /// Send work to a separate thread.
     template <typename T>
-    void Record(T&& command) {
+        requires std::is_invocable_v<T, vk::CommandBuffer, vk::CommandBuffer>
+    void RecordWithUploadBuffer(T&& command) {
         if (chunk->Record(command)) {
             return;
         }
@@ -88,6 +89,15 @@ public:
         (void)chunk->Record(command);
     }
 
+    template <typename T>
+        requires std::is_invocable_v<T, vk::CommandBuffer>
+    void Record(T&& c) {
+        this->RecordWithUploadBuffer(
+            [command = std::move(c)](vk::CommandBuffer cmdbuf, vk::CommandBuffer) {
+                command(cmdbuf);
+            });
+    }
+
     /// Returns the current command buffer tick.
     [[nodiscard]] u64 CurrentTick() const noexcept {
         return master_semaphore->CurrentTick();
@@ -119,7 +129,7 @@ private:
     public:
         virtual ~Command() = default;
 
-        virtual void Execute(vk::CommandBuffer cmdbuf) const = 0;
+        virtual void Execute(vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf) const = 0;
 
         Command* GetNext() const {
             return next;
@@ -142,8 +152,8 @@ private:
         TypedCommand(TypedCommand&&) = delete;
         TypedCommand& operator=(TypedCommand&&) = delete;
 
-        void Execute(vk::CommandBuffer cmdbuf) const override {
-            command(cmdbuf);
+        void Execute(vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf) const override {
+            command(cmdbuf, upload_cmdbuf);
         }
 
     private:
@@ -152,7 +162,7 @@ private:
 
     class CommandChunk final {
     public:
-        void ExecuteAll(vk::CommandBuffer cmdbuf);
+        void ExecuteAll(vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf);
 
         template <typename T>
         bool Record(T& command) {
@@ -228,6 +238,7 @@ private:
     VideoCommon::QueryCacheBase<QueryCacheParams>* query_cache = nullptr;
 
     vk::CommandBuffer current_cmdbuf;
+    vk::CommandBuffer current_upload_cmdbuf;
 
     std::unique_ptr<CommandChunk> chunk;
     std::function<void()> on_submit;
diff --git a/src/video_core/renderer_vulkan/vk_smaa.cpp b/src/video_core/renderer_vulkan/vk_smaa.cpp
index 5efd7d66e..70644ea82 100644
--- a/src/video_core/renderer_vulkan/vk_smaa.cpp
+++ b/src/video_core/renderer_vulkan/vk_smaa.cpp
@@ -672,7 +672,7 @@ void SMAA::UploadImages(Scheduler& scheduler) {
     UploadImage(m_device, m_allocator, scheduler, m_static_images[Search], search_extent,
                 VK_FORMAT_R8_UNORM, ARRAY_TO_SPAN(searchTexBytes));
 
-    scheduler.Record([&](vk::CommandBuffer& cmdbuf) {
+    scheduler.Record([&](vk::CommandBuffer cmdbuf) {
         for (auto& images : m_dynamic_images) {
             for (size_t i = 0; i < MaxDynamicImage; i++) {
                 ClearColorImage(cmdbuf, *images.images[i]);
@@ -707,7 +707,7 @@ VkImageView SMAA::Draw(Scheduler& scheduler, size_t image_index, VkImage source_
     UpdateDescriptorSets(source_image_view, image_index);
 
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([=, this](vk::CommandBuffer& cmdbuf) {
+    scheduler.Record([=, this](vk::CommandBuffer cmdbuf) {
         TransitionImageLayout(cmdbuf, source_image, VK_IMAGE_LAYOUT_GENERAL);
         TransitionImageLayout(cmdbuf, edges_image, VK_IMAGE_LAYOUT_GENERAL);
         BeginRenderPass(cmdbuf, m_renderpasses[EdgeDetection], edge_detection_framebuffer,
diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h
index d3deb9072..f63a20327 100644
--- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h
+++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.h
@@ -36,6 +36,10 @@ public:
     StagingBufferRef Request(size_t size, MemoryUsage usage, bool deferred = false);
     void FreeDeferred(StagingBufferRef& ref);
 
+    [[nodiscard]] VkBuffer StreamBuf() const noexcept {
+        return *stream_buffer;
+    }
+
     void TickFrame();
 
 private:
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index de34f6d49..5dbec2e62 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -1785,8 +1785,22 @@ ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::ImageInfo& info,
     : VideoCommon::ImageViewBase{info, view_info, gpu_addr_},
       buffer_size{VideoCommon::CalculateGuestSizeInBytes(info)} {}
 
-ImageView::ImageView(TextureCacheRuntime&, const VideoCommon::NullImageViewParams& params)
-    : VideoCommon::ImageViewBase{params} {}
+ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::NullImageViewParams& params)
+    : VideoCommon::ImageViewBase{params}, device{&runtime.device} {
+    if (device->HasNullDescriptor()) {
+        return;
+    }
+
+    // Handle fallback for devices without nullDescriptor
+    ImageInfo info{};
+    info.format = PixelFormat::A8B8G8R8_UNORM;
+
+    null_image = MakeImage(*device, runtime.memory_allocator, info, {});
+    image_handle = *null_image;
+    for (u32 i = 0; i < Shader::NUM_TEXTURE_TYPES; i++) {
+        image_views[i] = MakeView(VK_FORMAT_A8B8G8R8_UNORM_PACK32, VK_IMAGE_ASPECT_COLOR_BIT);
+    }
+}
 
 ImageView::~ImageView() = default;
 
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h
index 7a0807709..edf5d7635 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.h
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.h
@@ -267,6 +267,7 @@ private:
     vk::ImageView depth_view;
     vk::ImageView stencil_view;
     vk::ImageView color_view;
+    vk::Image null_image;
     VkImage image_handle = VK_NULL_HANDLE;
     VkImageView render_target = VK_NULL_HANDLE;
     VkSampleCountFlagBits samples = VK_SAMPLE_COUNT_1_BIT;