"Merge Tagged PR 6598"

.ci/templates/build-msvc.yml

@@ -8,7 +8,7 @@ steps:
   displayName: 'Install vulkan-sdk'
 - script: python -m pip install --upgrade pip conan
   displayName: 'Install conan'
-- script: refreshenv && mkdir build && cd build && cmake -G "Visual Studio 16 2019" -A x64 -DYUZU_USE_BUNDLED_QT=1 -DYUZU_USE_BUNDLED_SDL2=1 -DYUZU_USE_QT_WEB_ENGINE=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${COMPAT} -DUSE_DISCORD_PRESENCE=ON -DENABLE_QT_TRANSLATION=ON -DDISPLAY_VERSION=${{ parameters['version'] }} -DCMAKE_BUILD_TYPE=Release .. && cd ..
+- script: refreshenv && mkdir build && cd build && cmake -E env CXXFLAGS="/Gw /GA /Gr /Ob2" cmake -G "Visual Studio 16 2019" -A x64 -DCMAKE_INTERPROCEDURAL_OPTIMIZATION=ON -DCMAKE_POLICY_DEFAULT_CMP0069=NEW -DYUZU_USE_BUNDLED_QT=1 -DYUZU_USE_BUNDLED_SDL2=1 -DYUZU_USE_QT_WEB_ENGINE=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${COMPAT} -DUSE_DISCORD_PRESENCE=ON -DENABLE_QT_TRANSLATION=ON -DDISPLAY_VERSION=${{ parameters['version'] }} -DCMAKE_BUILD_TYPE=Release .. && cd ..
   displayName: 'Configure CMake'
 - task: MSBuild@1
   displayName: 'Build'

src/CMakeLists.txt

@@ -37,7 +37,8 @@ if (MSVC)
     add_compile_options(
         /MP
         /Zi
-        /Zm200
+        /Zm300
+        /Zf
         /Zo
         /permissive-
         /EHsc
@@ -81,7 +82,7 @@ if (MSVC)
     add_compile_options("$<$<CONFIG:Release>:/GS->")
 
     set(CMAKE_EXE_LINKER_FLAGS_DEBUG   "/DEBUG /MANIFEST:NO" CACHE STRING "" FORCE)
-    set(CMAKE_EXE_LINKER_FLAGS_RELEASE "/DEBUG /MANIFEST:NO /INCREMENTAL:NO /OPT:REF,ICF" CACHE STRING "" FORCE)
+    set(CMAKE_EXE_LINKER_FLAGS_RELEASE "${CMAKE_EXE_LINKER_FLAGS_RELEASE} /DEBUG /MANIFEST:NO /INCREMENTAL:NO /OPT:REF,ICF" CACHE STRING "" FORCE)
 else()
     add_compile_options(
         -Wall

src/audio_core/stream.cpp

@@ -87,6 +87,14 @@ static void VolumeAdjustSamples(std::vector<s16>& samples, float game_volume) {
 }
 
 void Stream::PlayNextBuffer(std::chrono::nanoseconds ns_late) {
+    auto now = std::chrono::steady_clock::now();
+    auto duration = now.time_since_epoch();
+    auto nanoseconds = std::chrono::duration_cast<std::chrono::nanoseconds>(duration);
+
+    if (nanoseconds > expected_cb_time) {
+        ns_late = nanoseconds - expected_cb_time;
+    }
+
     if (!IsPlaying()) {
         // Ensure we are in playing state before playing the next buffer
         sink_stream.Flush();
@@ -121,6 +129,7 @@ void Stream::PlayNextBuffer(std::chrono::nanoseconds ns_late) {
         ns_late = {};
     }
 
+    expected_cb_time = nanoseconds + (buffer_release_ns - ns_late);
     core_timing.ScheduleEvent(buffer_release_ns - ns_late, release_event, {});
 }
 

src/audio_core/stream.h

@@ -117,13 +117,14 @@ private:
     ReleaseCallback release_callback; ///< Buffer release callback for the stream
     State state{State::Stopped};      ///< Playback state of the stream
     std::shared_ptr<Core::Timing::EventType>
-        release_event;                      ///< Core timing release event for the stream
-    BufferPtr active_buffer;                ///< Actively playing buffer in the stream
-    std::queue<BufferPtr> queued_buffers;   ///< Buffers queued to be played in the stream
-    std::queue<BufferPtr> released_buffers; ///< Buffers recently released from the stream
-    SinkStream& sink_stream;                ///< Output sink for the stream
-    Core::Timing::CoreTiming& core_timing;  ///< Core timing instance.
-    std::string name;                       ///< Name of the stream, must be unique
+        release_event;                              ///< Core timing release event for the stream
+    BufferPtr active_buffer;                        ///< Actively playing buffer in the stream
+    std::queue<BufferPtr> queued_buffers;           ///< Buffers queued to be played in the stream
+    std::queue<BufferPtr> released_buffers;         ///< Buffers recently released from the stream
+    SinkStream& sink_stream;                        ///< Output sink for the stream
+    Core::Timing::CoreTiming& core_timing;          ///< Core timing instance.
+    std::string name;                               ///< Name of the stream, must be unique
+    std::chrono::nanoseconds expected_cb_time = {}; ///< Estimated time of next callback
 };
 
 using StreamPtr = std::shared_ptr<Stream>;

src/common/assert.h

@@ -32,14 +32,14 @@ assert_noinline_call(const Fn& fn) {
 
 #define ASSERT(_a_)                                                                                \
     do                                                                                             \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             assert_noinline_call([] { LOG_CRITICAL(Debug, "Assertion Failed!"); });                \
         }                                                                                          \
     while (0)
 
 #define ASSERT_MSG(_a_, ...)                                                                       \
     do                                                                                             \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             assert_noinline_call([&] { LOG_CRITICAL(Debug, "Assertion Failed!\n" __VA_ARGS__); }); \
         }                                                                                          \
     while (0)
@@ -70,7 +70,7 @@ assert_noinline_call(const Fn& fn) {
 #define ASSERT_OR_EXECUTE(_a_, _b_)                                                                \
     do {                                                                                           \
         ASSERT(_a_);                                                                               \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             _b_                                                                                    \
         }                                                                                          \
     } while (0)
@@ -79,7 +79,7 @@ assert_noinline_call(const Fn& fn) {
 #define ASSERT_OR_EXECUTE_MSG(_a_, _b_, ...)                                                       \
     do {                                                                                           \
         ASSERT_MSG(_a_, __VA_ARGS__);                                                              \
-        if (!(_a_)) {                                                                              \
+        if (!(_a_)) [[unlikely]] {                                                                 \
             _b_                                                                                    \
         }                                                                                          \
     } while (0)

src/video_core/buffer_cache/buffer_cache.h

@@ -1469,19 +1469,29 @@ typename BufferCache<P>::OverlapResult BufferCache<P>::ResolveOverlaps(VAddr cpu
         overlap_ids.push_back(overlap_id);
         overlap.Pick();
         const VAddr overlap_cpu_addr = overlap.CpuAddr();
+        bool goes_left = false;
         if (overlap_cpu_addr < begin) {
+            goes_left = true;
             cpu_addr = begin = overlap_cpu_addr;
         }
-        end = std::max(end, overlap_cpu_addr + overlap.SizeBytes());
-
+        const VAddr overlap_end = overlap_cpu_addr + overlap.SizeBytes();
+        bool goes_right = false;
+        if (overlap_end > end) {
+            goes_right = true;
+            end = overlap_end;
+        }
         stream_score += overlap.StreamScore();
         if (stream_score > STREAM_LEAP_THRESHOLD && !has_stream_leap) {
             // When this memory region has been joined a bunch of times, we assume it's being used
             // as a stream buffer. Increase the size to skip constantly recreating buffers.
             has_stream_leap = true;
-            begin -= PAGE_SIZE * 256;
-            cpu_addr = begin;
-            end += PAGE_SIZE * 256;
+            if (goes_right) {
+                begin -= PAGE_SIZE * 256;
+                cpu_addr = begin;
+            }
+            if (goes_left) {
+                end += PAGE_SIZE * 256;
+            }
         }
     }
     return OverlapResult{

src/video_core/macro/macro_hle.cpp

@@ -4,6 +4,8 @@
 
 #include <array>
 #include <vector>
+#include "common/scope_exit.h"
+#include "video_core/dirty_flags.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/macro/macro.h"
 #include "video_core/macro/macro_hle.h"
@@ -59,6 +61,7 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     maxwell3d.regs.index_array.first = parameters[3];
     maxwell3d.regs.reg_array[0x446] = element_base; // vertex id base?
     maxwell3d.regs.index_array.count = parameters[1];
+    maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
     maxwell3d.regs.vb_element_base = element_base;
     maxwell3d.regs.vb_base_instance = base_instance;
     maxwell3d.mme_draw.instance_count = instance_count;
@@ -81,10 +84,67 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
 }
 
-constexpr std::array<std::pair<u64, HLEFunction>, 3> hle_funcs{{
+// Multidraw Indirect
+void HLE_3f5e74b9c9a50164(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters) {
+    SCOPE_EXIT({
+        // Clean everything.
+        maxwell3d.regs.reg_array[0x446] = 0x0; // vertex id base?
+        maxwell3d.regs.index_array.count = 0;
+        maxwell3d.regs.vb_element_base = 0x0;
+        maxwell3d.regs.vb_base_instance = 0x0;
+        maxwell3d.mme_draw.instance_count = 0;
+        maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+        maxwell3d.CallMethodFromMME(0x8e4, 0x0);
+        maxwell3d.CallMethodFromMME(0x8e5, 0x0);
+        maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+        maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+    });
+    const u32 start_indirect = parameters[0];
+    const u32 end_indirect = parameters[1];
+    if (start_indirect >= end_indirect) {
+        // Nothing to do.
+        return;
+    }
+    const auto topology =
+        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[2]);
+    maxwell3d.regs.draw.topology.Assign(topology);
+    const u32 padding = parameters[3];
+    const std::size_t max_draws = parameters[4];
+
+    const u32 indirect_words = 5 + padding;
+    const std::size_t first_draw = start_indirect;
+    const std::size_t effective_draws = end_indirect - start_indirect;
+    const std::size_t last_draw = start_indirect + std::min(effective_draws, max_draws);
+
+    for (std::size_t index = first_draw; index < last_draw; index++) {
+        const std::size_t base = index * indirect_words + 5;
+        const u32 num_vertices = parameters[base];
+        const u32 instance_count = parameters[base + 1];
+        const u32 first_index = parameters[base + 2];
+        const u32 base_vertex = parameters[base + 3];
+        const u32 base_instance = parameters[base + 4];
+        maxwell3d.regs.index_array.first = first_index;
+        maxwell3d.regs.reg_array[0x446] = base_vertex;
+        maxwell3d.regs.index_array.count = num_vertices;
+        maxwell3d.regs.vb_element_base = base_vertex;
+        maxwell3d.regs.vb_base_instance = base_instance;
+        maxwell3d.mme_draw.instance_count = instance_count;
+        maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+        maxwell3d.CallMethodFromMME(0x8e4, base_vertex);
+        maxwell3d.CallMethodFromMME(0x8e5, base_instance);
+        maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+        if (maxwell3d.ShouldExecute()) {
+            maxwell3d.Rasterizer().Draw(true, true);
+        }
+        maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+    }
+}
+
+constexpr std::array<std::pair<u64, HLEFunction>, 4> hle_funcs{{
     {0x771BB18C62444DA0, &HLE_771BB18C62444DA0},
     {0x0D61FC9FAAC9FCAD, &HLE_0D61FC9FAAC9FCAD},
     {0x0217920100488FF7, &HLE_0217920100488FF7},
+    {0x3f5e74b9c9a50164, &HLE_3f5e74b9c9a50164},
 }};
 
 class HLEMacroImpl final : public CachedMacro {
@@ -100,6 +160,7 @@ private:
     Engines::Maxwell3D& maxwell3d;
     HLEFunction func;
 };
+
 } // Anonymous namespace
 
 HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d_) : maxwell3d{maxwell3d_} {}

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -554,10 +554,12 @@ void CopyBufferToImage(vk::CommandBuffer cmdbuf, VkBuffer src_buffer, VkImage im
     };
 }
 
-[[nodiscard]] bool IsFormatFlipped(PixelFormat format) {
+[[nodiscard]] bool IsFormatFlipped(PixelFormat format, bool emulate_bgr565) {
     switch (format) {
     case PixelFormat::A1B5G5R5_UNORM:
         return true;
+    case PixelFormat::B5G6R5_UNORM:
+        return emulate_bgr565;
     default:
         return false;
     }
@@ -1488,7 +1490,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
     };
     if (!info.IsRenderTarget()) {
         swizzle = info.Swizzle();
-        if (IsFormatFlipped(format)) {
+        if (IsFormatFlipped(format, device->MustEmulateBGR565())) {
             std::ranges::transform(swizzle, swizzle.begin(), SwapBlueRed);
         }
         if ((aspect_mask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) != 0) {

src/video_core/vulkan_common/vulkan_device.cpp

@@ -39,6 +39,11 @@ constexpr std::array DEPTH16_UNORM_STENCIL8_UINT{
     VK_FORMAT_D32_SFLOAT_S8_UINT,
     VK_FORMAT_UNDEFINED,
 };
+
+constexpr std::array B5G6R5_UNORM_PACK16{
+    VK_FORMAT_R5G6B5_UNORM_PACK16,
+    VK_FORMAT_UNDEFINED,
+};
 } // namespace Alternatives
 
 enum class NvidiaArchitecture {
@@ -87,6 +92,8 @@ constexpr const VkFormat* GetFormatAlternatives(VkFormat format) {
         return Alternatives::DEPTH24_UNORM_STENCIL8_UINT.data();
     case VK_FORMAT_D16_UNORM_S8_UINT:
         return Alternatives::DEPTH16_UNORM_STENCIL8_UINT.data();
+    case VK_FORMAT_B5G6R5_UNORM_PACK16:
+        return Alternatives::B5G6R5_UNORM_PACK16.data();
     default:
         return nullptr;
     }
@@ -639,6 +646,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
     }
 
     const bool is_intel_windows = driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS;
+    const bool is_intel_anv = driver_id == VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA;
     if (ext_vertex_input_dynamic_state && is_intel_windows) {
         LOG_WARNING(Render_Vulkan, "Blacklisting Intel for VK_EXT_vertex_input_dynamic_state");
         ext_vertex_input_dynamic_state = false;
@@ -652,6 +660,10 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
         LOG_WARNING(Render_Vulkan, "Intel proprietary drivers do not support MSAA image blits");
         cant_blit_msaa = true;
     }
+    if (is_intel_anv) {
+        LOG_WARNING(Render_Vulkan, "ANV driver does not support native BGR format");
+        must_emulate_bgr565 = true;
+    }
 
     supports_d24_depth =
         IsFormatSupported(VK_FORMAT_D24_UNORM_S8_UINT,

src/video_core/vulkan_common/vulkan_device.h

@@ -354,6 +354,10 @@ public:
         return cant_blit_msaa;
     }
 
+    bool MustEmulateBGR565() const {
+        return must_emulate_bgr565;
+    }
+
 private:
     /// Checks if the physical device is suitable.
     void CheckSuitability(bool requires_swapchain) const;
@@ -448,6 +452,7 @@ private:
     bool has_nsight_graphics{};             ///< Has Nsight Graphics attached
     bool supports_d24_depth{};              ///< Supports D24 depth buffers.
     bool cant_blit_msaa{};                  ///< Does not support MSAA<->MSAA blitting.
+    bool must_emulate_bgr565{};             ///< Emulates BGR565 by swizzling RGB565 format.
 
     // Telemetry parameters
     std::string vendor_name;                       ///< Device's driver name.