early-access version 1936

CMakeLists.txt

@@ -496,7 +496,7 @@ endif()
 # Ensure libusb is properly configured (based on dolphin libusb include)
 if(NOT APPLE AND NOT YUZU_USE_BUNDLED_LIBUSB)
     include(FindPkgConfig)
-    if (PKG_CONFIG_FOUND)
+    if (PKG_CONFIG_FOUND AND NOT CMAKE_SYSTEM_NAME MATCHES "DragonFly|FreeBSD")
         pkg_check_modules(LIBUSB QUIET libusb-1.0>=1.0.24)
     else()
         find_package(LibUSB)

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 1935.
+This is the source code for early-access 1936.
 
 ## Legal Notice
 

src/shader_recompiler/frontend/ir/value.h

@@ -57,6 +57,7 @@ public:
 
     [[nodiscard]] IR::Inst* Inst() const;
     [[nodiscard]] IR::Inst* InstRecursive() const;
+    [[nodiscard]] IR::Inst* TryInstRecursive() const;
     [[nodiscard]] IR::Value Resolve() const;
     [[nodiscard]] IR::Reg Reg() const;
     [[nodiscard]] IR::Pred Pred() const;
@@ -308,6 +309,13 @@ inline IR::Inst* Value::InstRecursive() const {
     return inst;
 }
 
+inline IR::Inst* Value::TryInstRecursive() const {
+    if (IsIdentity()) {
+        return inst->Arg(0).TryInstRecursive();
+    }
+    return type == Type::Opaque ? inst : nullptr;
+}
+
 inline IR::Value Value::Resolve() const {
     if (IsIdentity()) {
         return inst->Arg(0).Resolve();

src/shader_recompiler/ir_opt/constant_propagation_pass.cpp

@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
+#include <functional>
 #include <tuple>
 #include <type_traits>
 
@@ -88,6 +89,26 @@ bool FoldWhenAllImmediates(IR::Inst& inst, Func&& func) {
     return true;
 }
 
+/// Return true when all values in a range are equal
+template <typename Range>
+bool AreEqual(const Range& range) {
+    auto resolver{[](const auto& value) { return value.Resolve(); }};
+    auto equal{[](const IR::Value& lhs, const IR::Value& rhs) {
+        if (lhs == rhs) {
+            return true;
+        }
+        // Not equal, but try to match if they read the same constant buffer
+        if (!lhs.IsImmediate() && !rhs.IsImmediate() &&
+            lhs.Inst()->GetOpcode() == IR::Opcode::GetCbufU32 &&
+            rhs.Inst()->GetOpcode() == IR::Opcode::GetCbufU32 &&
+            lhs.Inst()->Arg(0) == rhs.Inst()->Arg(0) && lhs.Inst()->Arg(1) == rhs.Inst()->Arg(1)) {
+            return true;
+        }
+        return false;
+    }};
+    return std::ranges::adjacent_find(range, std::not_fn(equal), resolver) == std::end(range);
+}
+
 void FoldGetRegister(IR::Inst& inst) {
     if (inst.Arg(0).Reg() == IR::Reg::RZ) {
         inst.ReplaceUsesWith(IR::Value{u32{0}});
@@ -100,6 +121,157 @@ void FoldGetPred(IR::Inst& inst) {
     }
 }
 
+/// Replaces the XMAD pattern generated by an integer FMA
+bool FoldXmadMultiplyAdd(IR::Block& block, IR::Inst& inst) {
+    /*
+     * We are looking for this specific pattern:
+     *   %6      = BitFieldUExtract %op_b, #0, #16
+     *   %7      = BitFieldUExtract %op_a', #16, #16
+     *   %8      = IMul32 %6, %7
+     *   %10     = BitFieldUExtract %op_a', #0, #16
+     *   %11     = BitFieldInsert %8, %10, #16, #16
+     *   %15     = BitFieldUExtract %op_b, #0, #16
+     *   %16     = BitFieldUExtract %op_a, #0, #16
+     *   %17     = IMul32 %15, %16
+     *   %18     = IAdd32 %17, %op_c
+     *   %22     = BitFieldUExtract %op_b, #16, #16
+     *   %23     = BitFieldUExtract %11, #16, #16
+     *   %24     = IMul32 %22, %23
+     *   %25     = ShiftLeftLogical32 %24, #16
+     *   %26     = ShiftLeftLogical32 %11, #16
+     *   %27     = IAdd32 %26, %18
+     *   %result = IAdd32 %25, %27
+     *
+     * And replace it with:
+     *  %temp   = IMul32 %op_a, %op_b
+     *  %result = IAdd32 %temp, %op_c
+     *
+     * This optimization has been proven safe by Nvidia's compiler logic being reversed.
+     * (If Nvidia generates this code from 'fma(a, b, c)', we can do the same in the reverse order.)
+     */
+    const IR::Value zero{0u};
+    const IR::Value sixteen{16u};
+    IR::Inst* const _25{inst.Arg(0).TryInstRecursive()};
+    IR::Inst* const _27{inst.Arg(1).TryInstRecursive()};
+    if (!_25 || !_27) {
+        return false;
+    }
+    if (_27->GetOpcode() != IR::Opcode::IAdd32) {
+        return false;
+    }
+    if (_25->GetOpcode() != IR::Opcode::ShiftLeftLogical32 || _25->Arg(1) != sixteen) {
+        return false;
+    }
+    IR::Inst* const _24{_25->Arg(0).TryInstRecursive()};
+    if (!_24 || _24->GetOpcode() != IR::Opcode::IMul32) {
+        return false;
+    }
+    IR::Inst* const _22{_24->Arg(0).TryInstRecursive()};
+    IR::Inst* const _23{_24->Arg(1).TryInstRecursive()};
+    if (!_22 || !_23) {
+        return false;
+    }
+    if (_22->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_23->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_22->Arg(1) != sixteen || _22->Arg(2) != sixteen) {
+        return false;
+    }
+    if (_23->Arg(1) != sixteen || _23->Arg(2) != sixteen) {
+        return false;
+    }
+    IR::Inst* const _11{_23->Arg(0).TryInstRecursive()};
+    if (!_11 || _11->GetOpcode() != IR::Opcode::BitFieldInsert) {
+        return false;
+    }
+    if (_11->Arg(2) != sixteen || _11->Arg(3) != sixteen) {
+        return false;
+    }
+    IR::Inst* const _8{_11->Arg(0).TryInstRecursive()};
+    IR::Inst* const _10{_11->Arg(1).TryInstRecursive()};
+    if (!_8 || !_10) {
+        return false;
+    }
+    if (_8->GetOpcode() != IR::Opcode::IMul32) {
+        return false;
+    }
+    if (_10->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    IR::Inst* const _6{_8->Arg(0).TryInstRecursive()};
+    IR::Inst* const _7{_8->Arg(1).TryInstRecursive()};
+    if (!_6 || !_7) {
+        return false;
+    }
+    if (_6->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_7->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_6->Arg(1) != zero || _6->Arg(2) != sixteen) {
+        return false;
+    }
+    if (_7->Arg(1) != sixteen || _7->Arg(2) != sixteen) {
+        return false;
+    }
+    IR::Inst* const _26{_27->Arg(0).TryInstRecursive()};
+    IR::Inst* const _18{_27->Arg(1).TryInstRecursive()};
+    if (!_26 || !_18) {
+        return false;
+    }
+    if (_26->GetOpcode() != IR::Opcode::ShiftLeftLogical32 || _26->Arg(1) != sixteen) {
+        return false;
+    }
+    if (_26->Arg(0).InstRecursive() != _11) {
+        return false;
+    }
+    if (_18->GetOpcode() != IR::Opcode::IAdd32) {
+        return false;
+    }
+    IR::Inst* const _17{_18->Arg(0).TryInstRecursive()};
+    if (!_17 || _17->GetOpcode() != IR::Opcode::IMul32) {
+        return false;
+    }
+    IR::Inst* const _15{_17->Arg(0).TryInstRecursive()};
+    IR::Inst* const _16{_17->Arg(1).TryInstRecursive()};
+    if (!_15 || !_16) {
+        return false;
+    }
+    if (_15->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_16->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_15->Arg(1) != zero || _16->Arg(1) != zero || _10->Arg(1) != zero) {
+        return false;
+    }
+    if (_15->Arg(2) != sixteen || _16->Arg(2) != sixteen || _10->Arg(2) != sixteen) {
+        return false;
+    }
+    const std::array<IR::Value, 3> op_as{
+        _7->Arg(0).Resolve(),
+        _16->Arg(0).Resolve(),
+        _10->Arg(0).Resolve(),
+    };
+    const std::array<IR::Value, 3> op_bs{
+        _22->Arg(0).Resolve(),
+        _6->Arg(0).Resolve(),
+        _15->Arg(0).Resolve(),
+    };
+    const IR::U32 op_c{_18->Arg(1)};
+    if (!AreEqual(op_as) || !AreEqual(op_bs)) {
+        return false;
+    }
+    IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
+    inst.ReplaceUsesWith(ir.IAdd(ir.IMul(IR::U32{op_as[0]}, IR::U32{op_bs[1]}), op_c));
+    return true;
+}
+
 /// Replaces the pattern generated by two XMAD multiplications
 bool FoldXmadMultiply(IR::Block& block, IR::Inst& inst) {
     /*
@@ -116,33 +288,31 @@ bool FoldXmadMultiply(IR::Block& block, IR::Inst& inst) {
      *
      * This optimization has been proven safe by LLVM and MSVC.
      */
-    const IR::Value lhs_arg{inst.Arg(0)};
+    IR::Inst* const lhs_shl{inst.Arg(0).TryInstRecursive()};
-    const IR::Value rhs_arg{inst.Arg(1)};
+    IR::Inst* const rhs_mul{inst.Arg(1).TryInstRecursive()};
-    if (lhs_arg.IsImmediate() || rhs_arg.IsImmediate()) {
+    if (!lhs_shl || !rhs_mul) {
         return false;
     }
-    IR::Inst* const lhs_shl{lhs_arg.InstRecursive()};
     if (lhs_shl->GetOpcode() != IR::Opcode::ShiftLeftLogical32 ||
         lhs_shl->Arg(1) != IR::Value{16U}) {
         return false;
     }
-    if (lhs_shl->Arg(0).IsImmediate()) {
+    IR::Inst* const lhs_mul{lhs_shl->Arg(0).TryInstRecursive()};
+    if (!lhs_mul) {
         return false;
     }
-    IR::Inst* const lhs_mul{lhs_shl->Arg(0).InstRecursive()};
-    IR::Inst* const rhs_mul{rhs_arg.InstRecursive()};
     if (lhs_mul->GetOpcode() != IR::Opcode::IMul32 || rhs_mul->GetOpcode() != IR::Opcode::IMul32) {
         return false;
     }
-    if (lhs_mul->Arg(1).Resolve() != rhs_mul->Arg(1).Resolve()) {
-        return false;
-    }
     const IR::U32 factor_b{lhs_mul->Arg(1)};
-    if (lhs_mul->Arg(0).IsImmediate() || rhs_mul->Arg(0).IsImmediate()) {
+    if (factor_b.Resolve() != rhs_mul->Arg(1).Resolve()) {
+        return false;
+    }
+    IR::Inst* const lhs_bfe{lhs_mul->Arg(0).TryInstRecursive()};
+    IR::Inst* const rhs_bfe{rhs_mul->Arg(0).TryInstRecursive()};
+    if (!lhs_bfe || !rhs_bfe) {
         return false;
     }
-    IR::Inst* const lhs_bfe{lhs_mul->Arg(0).InstRecursive()};
-    IR::Inst* const rhs_bfe{rhs_mul->Arg(0).InstRecursive()};
     if (lhs_bfe->GetOpcode() != IR::Opcode::BitFieldUExtract) {
         return false;
     }
@@ -155,10 +325,10 @@ bool FoldXmadMultiply(IR::Block& block, IR::Inst& inst) {
     if (rhs_bfe->Arg(1) != IR::Value{0U} || rhs_bfe->Arg(2) != IR::Value{16U}) {
         return false;
     }
-    if (lhs_bfe->Arg(0).Resolve() != rhs_bfe->Arg(0).Resolve()) {
+    const IR::U32 factor_a{lhs_bfe->Arg(0)};
+    if (factor_a.Resolve() != rhs_bfe->Arg(0).Resolve()) {
         return false;
     }
-    const IR::U32 factor_a{lhs_bfe->Arg(0)};
     IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
     inst.ReplaceUsesWith(ir.IMul(factor_a, factor_b));
     return true;
@@ -181,6 +351,9 @@ void FoldAdd(IR::Block& block, IR::Inst& inst) {
         if (FoldXmadMultiply(block, inst)) {
             return;
         }
+        if (FoldXmadMultiplyAdd(block, inst)) {
+            return;
+        }
     }
 }
 

src/video_core/renderer_base.cpp

@@ -27,7 +27,7 @@ void RendererBase::UpdateCurrentFramebufferLayout() {
     render_window.UpdateCurrentFramebufferLayout(layout.width, layout.height);
 }
 
-void RendererBase::RequestScreenshot(void* data, std::function<void()> callback,
+void RendererBase::RequestScreenshot(void* data, std::function<void(bool)> callback,
                                      const Layout::FramebufferLayout& layout) {
     if (renderer_settings.screenshot_requested) {
         LOG_ERROR(Render, "A screenshot is already requested or in progress, ignoring the request");

src/video_core/renderer_base.h

@@ -24,7 +24,7 @@ struct RendererSettings {
     // Screenshot
     std::atomic<bool> screenshot_requested{false};
     void* screenshot_bits{};
-    std::function<void()> screenshot_complete_callback;
+    std::function<void(bool)> screenshot_complete_callback;
     Layout::FramebufferLayout screenshot_framebuffer_layout;
 };
 
@@ -80,7 +80,7 @@ public:
     void RefreshBaseSettings();
 
     /// Request a screenshot of the next frame
-    void RequestScreenshot(void* data, std::function<void()> callback,
+    void RequestScreenshot(void* data, std::function<void(bool)> callback,
                            const Layout::FramebufferLayout& layout);
 
 protected:

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -486,7 +486,7 @@ void RendererOpenGL::RenderScreenshot() {
     glBindFramebuffer(GL_READ_FRAMEBUFFER, old_read_fb);
     glBindFramebuffer(GL_DRAW_FRAMEBUFFER, old_draw_fb);
 
-    renderer_settings.screenshot_complete_callback();
+    renderer_settings.screenshot_complete_callback(true);
     renderer_settings.screenshot_requested = false;
 }
 

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -138,6 +138,7 @@ 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);
 
     bool has_been_recreated = false;
     const auto recreate_swapchain = [&] {
@@ -162,7 +163,7 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
     if (has_been_recreated) {
         blit_screen.Recreate();
     }
-    const VkSemaphore render_semaphore = blit_screen.Draw(*framebuffer, use_accelerated);
+    const VkSemaphore render_semaphore = blit_screen.DrawToSwapchain(*framebuffer, use_accelerated);
     scheduler.Flush(render_semaphore);
     scheduler.WaitWorker();
     swapchain.Present(render_semaphore);
@@ -193,4 +194,153 @@ void RendererVulkan::Report() const {
     telemetry_session.AddField(field, "GPU_Vulkan_Extensions", extensions);
 }
 
+void Vulkan::RendererVulkan::RenderScreenshot(const Tegra::FramebufferConfig& framebuffer,
+                                              bool use_accelerated) {
+    if (!renderer_settings.screenshot_requested) {
+        return;
+    }
+    const Layout::FramebufferLayout layout{renderer_settings.screenshot_framebuffer_layout};
+    vk::Image staging_image = device.GetLogical().CreateImage(VkImageCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT,
+        .imageType = VK_IMAGE_TYPE_2D,
+        .format = VK_FORMAT_B8G8R8A8_UNORM,
+        .extent =
+            {
+                .width = layout.width,
+                .height = layout.height,
+                .depth = 1,
+            },
+        .mipLevels = 1,
+        .arrayLayers = 1,
+        .samples = VK_SAMPLE_COUNT_1_BIT,
+        .tiling = VK_IMAGE_TILING_OPTIMAL,
+        .usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT |
+                 VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT,
+        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+        .queueFamilyIndexCount = 0,
+        .pQueueFamilyIndices = nullptr,
+        .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+    });
+    const auto image_commit = memory_allocator.Commit(staging_image, MemoryUsage::DeviceLocal);
+
+    const vk::ImageView dst_view = device.GetLogical().CreateImageView(VkImageViewCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .image = *staging_image,
+        .viewType = VK_IMAGE_VIEW_TYPE_2D,
+        .format = screen_info.is_srgb ? VK_FORMAT_B8G8R8A8_SRGB : VK_FORMAT_B8G8R8A8_UNORM,
+        .components{
+            .r = VK_COMPONENT_SWIZZLE_IDENTITY,
+            .g = VK_COMPONENT_SWIZZLE_IDENTITY,
+            .b = VK_COMPONENT_SWIZZLE_IDENTITY,
+            .a = VK_COMPONENT_SWIZZLE_IDENTITY,
+        },
+        .subresourceRange{
+            .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+            .baseMipLevel = 0,
+            .levelCount = 1,
+            .baseArrayLayer = 0,
+            .layerCount = VK_REMAINING_ARRAY_LAYERS,
+        },
+    });
+    const VkExtent2D render_area{.width = layout.width, .height = layout.height};
+    const vk::Framebuffer screenshot_fb = blit_screen.CreateFramebuffer(*dst_view, render_area);
+    // Since we're not rendering to the screen, ignore the render semaphore.
+    void(blit_screen.Draw(framebuffer, *screenshot_fb, layout, render_area, use_accelerated));
+
+    const auto buffer_size = static_cast<VkDeviceSize>(layout.width * layout.height * 4);
+    const VkBufferCreateInfo dst_buffer_info{
+        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .size = buffer_size,
+        .usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+        .queueFamilyIndexCount = 0,
+        .pQueueFamilyIndices = nullptr,
+    };
+    const vk::Buffer dst_buffer = device.GetLogical().CreateBuffer(dst_buffer_info);
+    MemoryCommit dst_buffer_memory = memory_allocator.Commit(dst_buffer, MemoryUsage::Download);
+
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([&](vk::CommandBuffer cmdbuf) {
+        const VkImageMemoryBarrier read_barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
+            .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = *staging_image,
+            .subresourceRange{
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        const VkImageMemoryBarrier image_write_barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = 0,
+            .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = *staging_image,
+            .subresourceRange{
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        static constexpr VkMemoryBarrier memory_write_barrier{
+            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
+        };
+        const VkBufferImageCopy copy{
+            .bufferOffset = 0,
+            .bufferRowLength = 0,
+            .bufferImageHeight = 0,
+            .imageSubresource{
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                .mipLevel = 0,
+                .baseArrayLayer = 0,
+                .layerCount = 1,
+            },
+            .imageOffset{.x = 0, .y = 0, .z = 0},
+            .imageExtent{
+                .width = layout.width,
+                .height = layout.height,
+                .depth = 1,
+            },
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
+                               0, read_barrier);
+        cmdbuf.CopyImageToBuffer(*staging_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, *dst_buffer,
+                                 copy);
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+                               0, memory_write_barrier, nullptr, image_write_barrier);
+    });
+    // Ensure the copy is fully completed before saving the screenshot
+    scheduler.Finish();
+
+    // Copy backing image data to the QImage screenshot buffer
+    const auto dst_memory_map = dst_buffer_memory.Map();
+    std::memcpy(renderer_settings.screenshot_bits, dst_memory_map.data(), dst_memory_map.size());
+    renderer_settings.screenshot_complete_callback(false);
+    renderer_settings.screenshot_requested = false;
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/renderer_vulkan.h

@@ -54,6 +54,8 @@ public:
 private:
     void Report() const;
 
+    void RenderScreenshot(const Tegra::FramebufferConfig& framebuffer, bool use_accelerated);
+
     Core::TelemetrySession& telemetry_session;
     Core::Memory::Memory& cpu_memory;
     Tegra::GPU& gpu;

src/video_core/renderer_vulkan/vk_blit_screen.cpp

@@ -130,7 +130,10 @@ void VKBlitScreen::Recreate() {
     CreateDynamicResources();
 }
 
-VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool use_accelerated) {
+VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
+                               const VkFramebuffer& host_framebuffer,
+                               const Layout::FramebufferLayout layout, VkExtent2D render_area,
+                               bool use_accelerated) {
     RefreshResources(framebuffer);
 
     // Finish any pending renderpass
@@ -145,8 +148,8 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
                         use_accelerated ? screen_info.image_view : *raw_image_views[image_index]);
 
     BufferData data;
-    SetUniformData(data, framebuffer);
+    SetUniformData(data, layout);
-    SetVertexData(data, framebuffer);
+    SetVertexData(data, framebuffer, layout);
 
     const std::span<u8> mapped_span = buffer_commit.Map();
     std::memcpy(mapped_span.data(), &data, sizeof(data));
@@ -220,7 +223,8 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
                                    VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier);
         });
     }
-    scheduler.Record([this, image_index, size = swapchain.GetSize()](vk::CommandBuffer cmdbuf) {
+    scheduler.Record(
+        [this, host_framebuffer, image_index, size = render_area](vk::CommandBuffer cmdbuf) {
             const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f;
             const f32 bg_green = Settings::values.bg_green.GetValue() / 255.0f;
             const f32 bg_blue = Settings::values.bg_blue.GetValue() / 255.0f;
@@ -231,7 +235,7 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
                 .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
                 .pNext = nullptr,
                 .renderPass = *renderpass,
-            .framebuffer = *framebuffers[image_index],
+                .framebuffer = host_framebuffer,
                 .renderArea =
                     {
                         .offset = {0, 0},
@@ -266,6 +270,28 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
     return *semaphores[image_index];
 }
 
+VkSemaphore VKBlitScreen::DrawToSwapchain(const Tegra::FramebufferConfig& framebuffer,
+                                          bool use_accelerated) {
+    const std::size_t image_index = swapchain.GetImageIndex();
+    const VkExtent2D render_area = swapchain.GetSize();
+    const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout();
+    return Draw(framebuffer, *framebuffers[image_index], layout, render_area, use_accelerated);
+}
+
+vk::Framebuffer VKBlitScreen::CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent) {
+    return device.GetLogical().CreateFramebuffer(VkFramebufferCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .renderPass = *renderpass,
+        .attachmentCount = 1,
+        .pAttachments = &image_view,
+        .width = extent.width,
+        .height = extent.height,
+        .layers = 1,
+    });
+}
+
 void VKBlitScreen::CreateStaticResources() {
     CreateShaders();
     CreateSemaphores();
@@ -752,15 +778,13 @@ void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView imag
     device.GetLogical().UpdateDescriptorSets(std::array{ubo_write, sampler_write}, {});
 }
 
-void VKBlitScreen::SetUniformData(BufferData& data,
+void VKBlitScreen::SetUniformData(BufferData& data, const Layout::FramebufferLayout layout) const {
-                                  const Tegra::FramebufferConfig& framebuffer) const {
-    const auto& layout = render_window.GetFramebufferLayout();
     data.uniform.modelview_matrix =
         MakeOrthographicMatrix(static_cast<f32>(layout.width), static_cast<f32>(layout.height));
 }
 
-void VKBlitScreen::SetVertexData(BufferData& data,
+void VKBlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer,
-                                 const Tegra::FramebufferConfig& framebuffer) const {
+                                 const Layout::FramebufferLayout layout) const {
     const auto& framebuffer_transform_flags = framebuffer.transform_flags;
     const auto& framebuffer_crop_rect = framebuffer.crop_rect;
 
@@ -798,7 +822,7 @@ void VKBlitScreen::SetVertexData(BufferData& data,
                   static_cast<f32>(screen_info.height);
     }
 
-    const auto& screen = render_window.GetFramebufferLayout().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());

src/video_core/renderer_vulkan/vk_blit_screen.h

@@ -56,8 +56,16 @@ public:
     void Recreate();
 
     [[nodiscard]] VkSemaphore Draw(const Tegra::FramebufferConfig& framebuffer,
+                                   const VkFramebuffer& host_framebuffer,
+                                   const Layout::FramebufferLayout layout, VkExtent2D render_area,
                                    bool use_accelerated);
 
+    [[nodiscard]] VkSemaphore DrawToSwapchain(const Tegra::FramebufferConfig& framebuffer,
+                                              bool use_accelerated);
+
+    [[nodiscard]] vk::Framebuffer CreateFramebuffer(const VkImageView& image_view,
+                                                    VkExtent2D extent);
+
 private:
     struct BufferData;
 
@@ -81,8 +89,9 @@ private:
     void CreateRawImages(const Tegra::FramebufferConfig& framebuffer);
 
     void UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const;
-    void SetUniformData(BufferData& data, const Tegra::FramebufferConfig& framebuffer) const;
+    void SetUniformData(BufferData& data, const Layout::FramebufferLayout layout) const;
-    void SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer) const;
+    void SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer,
+                       const Layout::FramebufferLayout layout) const;
 
     u64 CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const;
     u64 GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer,

src/yuzu/bootmanager.cpp

@@ -634,9 +634,9 @@ void GRenderWindow::CaptureScreenshot(u32 res_scale, const QString& screenshot_p
     screenshot_image = QImage(QSize(layout.width, layout.height), QImage::Format_RGB32);
     renderer.RequestScreenshot(
         screenshot_image.bits(),
-        [=, this] {
+        [=, this](bool invert_y) {
             const std::string std_screenshot_path = screenshot_path.toStdString();
-            if (screenshot_image.mirrored(false, true).save(screenshot_path)) {
+            if (screenshot_image.mirrored(false, invert_y).save(screenshot_path)) {
                 LOG_INFO(Frontend, "Screenshot saved to \"{}\"", std_screenshot_path);
             } else {
                 LOG_ERROR(Frontend, "Failed to save screenshot to \"{}\"", std_screenshot_path);