applets: software_keyboard: Block calling thread until text is available.

- Fixes a bug where guest thread is resumed too early (Little Town Hero).

CMakeLists.txt

@@ -350,6 +350,13 @@ function(create_target_directory_groups target_name)
     endforeach()
 endfunction()
 
+# Prevent boost from linking against libs when building
+add_definitions(-DBOOST_ERROR_CODE_HEADER_ONLY
+    -DBOOST_SYSTEM_NO_LIB
+    -DBOOST_DATE_TIME_NO_LIB
+    -DBOOST_REGEX_NO_LIB
+)
+
 enable_testing()
 add_subdirectory(externals)
 add_subdirectory(src)

CMakeModules/GenerateSCMRev.cmake

@@ -5,6 +5,10 @@ function(get_timestamp _var)
 endfunction()
 
 list(APPEND CMAKE_MODULE_PATH "${SRC_DIR}/externals/cmake-modules")
+
+# Find the package here with the known path so that the GetGit commands can find it as well
+find_package(Git QUIET PATHS "${GIT_EXECUTABLE}")
+
 # generate git/build information
 include(GetGitRevisionDescription)
 get_git_head_revision(GIT_REF_SPEC GIT_REV)

dist/qt_themes/qdarkstyle/style.qss

@@ -181,7 +181,7 @@ QMenu::icon {
 }
 
 QMenu::item {
-    padding: 5px 30px 5px 30px;
+    padding: 5px 16px 5px 40px;
     border: 1px solid transparent;
     /* reserve space for selection border */
 }
@@ -192,12 +192,13 @@ QMenu::item:selected {
 
 QMenu::separator {
     height: 2px;
-    background: lightblue;
+    background: #76797C;
     margin-left: 10px;
     margin-right: 5px;
 }
 
 QMenu::indicator {
+    margin: 0 -26px 0 8px;
     width: 18px;
     height: 18px;
 }
@@ -252,7 +253,7 @@ QWidget:disabled {
 }
 
 QAbstractItemView {
-    alternate-background-color: #31363b;
+    alternate-background-color: #2c2f32;
     color: #eff0f1;
     border: 1px solid #3A3939;
     border-radius: 2px;
@@ -577,8 +578,6 @@ QTreeView:hover {
 }
 
 QComboBox:on {
-    padding-top: 3px;
-    padding-left: 4px;
     selection-background-color: #4a4a4a;
 }
 
@@ -703,10 +702,10 @@ QTabBar::close-button:pressed {
 QTabBar::tab:top {
     color: #eff0f1;
     border: 1px solid #76797C;
-    border-bottom: 1px transparent black;
+    border-bottom: 2px transparent;
     background-color: #31363b;
-    padding: 5px;
-    min-width: 50px;
+    padding: 4px 16px 2px;
+    min-width: 38px;
     border-top-left-radius: 2px;
     border-top-right-radius: 2px;
 }
@@ -1078,7 +1077,7 @@ QListView::item:selected:active {
 }
 
 QHeaderView {
-    background-color: #31363b;
+    background-color: #403F3F;
     border: 1px transparent;
     border-radius: 0px;
     margin: 0px;
@@ -1086,30 +1085,32 @@ QHeaderView {
 }
 
 QHeaderView::section {
-    background-color: #31363b;
+    background-color: #232629;
     color: #eff0f1;
-    padding: 5px;
-    border: 1px solid #76797C;
+    padding: 0 5px;
+    border: 1px solid #403F3F;
+    border-bottom: 0;
     border-radius: 0px;
     text-align: center;
 }
 
 QHeaderView::section::vertical::first,
 QHeaderView::section::vertical::only-one {
-    border-top: 1px solid #76797C;
+    border-top: 1px solid #31363b;
 }
 
 QHeaderView::section::vertical {
     border-top: transparent;
 }
 
+QHeaderView::section::horizontal,
 QHeaderView::section::horizontal::first,
 QHeaderView::section::horizontal::only-one {
-    border-left: 1px solid #76797C;
+    border-left: transparent;
 }
 
-QHeaderView::section::horizontal {
-    border-left: transparent;
+QHeaderView::section::horizontal::last {
+    border-right: transparent;
 }
 
 QHeaderView::section:checked {

src/CMakeLists.txt

@@ -77,6 +77,15 @@ else()
             add_compile_options("-static")
         endif()
     endif()
+
+    if(CMAKE_SYSTEM_NAME STREQUAL "Linux" OR MINGW)
+        # GNU ar: Create thin archive files.
+        # Requires binutils-2.19 or later.
+        set(CMAKE_C_ARCHIVE_CREATE   "<CMAKE_AR> qcTP <TARGET> <LINK_FLAGS> <OBJECTS>")
+        set(CMAKE_C_ARCHIVE_APPEND   "<CMAKE_AR> qTP  <TARGET> <LINK_FLAGS> <OBJECTS>")
+        set(CMAKE_CXX_ARCHIVE_CREATE "<CMAKE_AR> qcTP <TARGET> <LINK_FLAGS> <OBJECTS>")
+        set(CMAKE_CXX_ARCHIVE_APPEND "<CMAKE_AR> qTP  <TARGET> <LINK_FLAGS> <OBJECTS>")
+    endif()
 endif()
 
 add_subdirectory(common)

src/common/CMakeLists.txt

@@ -15,6 +15,10 @@ endif ()
 if (DEFINED ENV{DISPLAYVERSION})
   set(DISPLAY_VERSION $ENV{DISPLAYVERSION})
 endif ()
+
+# Pass the path to git to the GenerateSCMRev.cmake as well
+find_package(Git QUIET)
+
 add_custom_command(OUTPUT scm_rev.cpp
     COMMAND ${CMAKE_COMMAND}
       -DSRC_DIR="${CMAKE_SOURCE_DIR}"
@@ -23,6 +27,7 @@ add_custom_command(OUTPUT scm_rev.cpp
       -DTITLE_BAR_FORMAT_RUNNING="${TITLE_BAR_FORMAT_RUNNING}"
       -DBUILD_TAG="${BUILD_TAG}"
       -DBUILD_ID="${DISPLAY_VERSION}"
+      -DGIT_EXECUTABLE="${GIT_EXECUTABLE}"
       -P "${CMAKE_SOURCE_DIR}/CMakeModules/GenerateSCMRev.cmake"
     DEPENDS
       # WARNING! It was too much work to try and make a common location for this list,

src/common/logging/backend.cpp

@@ -120,7 +120,7 @@ private:
             duration_cast<std::chrono::microseconds>(steady_clock::now() - time_origin);
         entry.log_class = log_class;
         entry.log_level = log_level;
-        entry.filename = Common::TrimSourcePath(filename);
+        entry.filename = filename;
         entry.line_num = line_nr;
         entry.function = function;
         entry.message = std::move(message);

src/common/logging/backend.h

@@ -23,7 +23,7 @@ struct Entry {
     std::chrono::microseconds timestamp;
     Class log_class;
     Level log_level;
-    std::string filename;
+    const char* filename;
     unsigned int line_num;
     std::string function;
     std::string message;

src/common/logging/log.h

@@ -9,6 +9,15 @@
 
 namespace Log {
 
+// trims up to and including the last of ../, ..\, src/, src\ in a string
+constexpr const char* TrimSourcePath(std::string_view source) {
+    const auto rfind = [source](const std::string_view match) {
+        return source.rfind(match) == source.npos ? 0 : (source.rfind(match) + match.size());
+    };
+    auto idx = std::max({rfind("src/"), rfind("src\\"), rfind("../"), rfind("..\\")});
+    return source.data() + idx;
+}
+
 /// Specifies the severity or level of detail of the log message.
 enum class Level : u8 {
     Trace,    ///< Extremely detailed and repetitive debugging information that is likely to
@@ -141,24 +150,24 @@ void FmtLogMessage(Class log_class, Level log_level, const char* filename, unsig
 
 #ifdef _DEBUG
 #define LOG_TRACE(log_class, ...)                                                                  \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Trace, __FILE__, __LINE__,         \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Trace,                             \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)
 #else
 #define LOG_TRACE(log_class, fmt, ...) (void(0))
 #endif
 
 #define LOG_DEBUG(log_class, ...)                                                                  \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Debug, __FILE__, __LINE__,         \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Debug,                             \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)
 #define LOG_INFO(log_class, ...)                                                                   \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Info, __FILE__, __LINE__,          \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Info,                              \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)
 #define LOG_WARNING(log_class, ...)                                                                \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Warning, __FILE__, __LINE__,       \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Warning,                           \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)
 #define LOG_ERROR(log_class, ...)                                                                  \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Error, __FILE__, __LINE__,         \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Error,                             \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)
 #define LOG_CRITICAL(log_class, ...)                                                               \
-    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Critical, __FILE__, __LINE__,      \
-                         __func__, __VA_ARGS__)
+    ::Log::FmtLogMessage(::Log::Class::log_class, ::Log::Level::Critical,                          \
+                         ::Log::TrimSourcePath(__FILE__), __LINE__, __func__, __VA_ARGS__)

src/common/string_util.cpp

@@ -223,26 +223,4 @@ std::u16string UTF16StringFromFixedZeroTerminatedBuffer(std::u16string_view buff
     return std::u16string(buffer.begin(), buffer.begin() + len);
 }
 
-const char* TrimSourcePath(const char* path, const char* root) {
-    const char* p = path;
-
-    while (*p != '\0') {
-        const char* next_slash = p;
-        while (*next_slash != '\0' && *next_slash != '/' && *next_slash != '\\') {
-            ++next_slash;
-        }
-
-        bool is_src = Common::ComparePartialString(p, next_slash, root);
-        p = next_slash;
-
-        if (*p != '\0') {
-            ++p;
-        }
-        if (is_src) {
-            path = p;
-        }
-    }
-    return path;
-}
-
 } // namespace Common

src/common/telemetry.cpp

@@ -44,20 +44,6 @@ template class Field<std::string>;
 template class Field<const char*>;
 template class Field<std::chrono::microseconds>;
 
-#ifdef ARCHITECTURE_x86_64
-static const char* CpuVendorToStr(Common::CPUVendor vendor) {
-    switch (vendor) {
-    case Common::CPUVendor::INTEL:
-        return "Intel";
-    case Common::CPUVendor::AMD:
-        return "Amd";
-    case Common::CPUVendor::OTHER:
-        return "Other";
-    }
-    UNREACHABLE();
-}
-#endif
-
 void AppendBuildInfo(FieldCollection& fc) {
     const bool is_git_dirty{std::strstr(Common::g_scm_desc, "dirty") != nullptr};
     fc.AddField(FieldType::App, "Git_IsDirty", is_git_dirty);
@@ -71,7 +57,6 @@ void AppendCPUInfo(FieldCollection& fc) {
 #ifdef ARCHITECTURE_x86_64
     fc.AddField(FieldType::UserSystem, "CPU_Model", Common::GetCPUCaps().cpu_string);
     fc.AddField(FieldType::UserSystem, "CPU_BrandString", Common::GetCPUCaps().brand_string);
-    fc.AddField(FieldType::UserSystem, "CPU_Vendor", CpuVendorToStr(Common::GetCPUCaps().vendor));
     fc.AddField(FieldType::UserSystem, "CPU_Extension_x64_AES", Common::GetCPUCaps().aes);
     fc.AddField(FieldType::UserSystem, "CPU_Extension_x64_AVX", Common::GetCPUCaps().avx);
     fc.AddField(FieldType::UserSystem, "CPU_Extension_x64_AVX2", Common::GetCPUCaps().avx2);

src/common/thread.h

@@ -28,6 +28,15 @@ public:
         is_set = false;
     }
 
+    template <class Duration>
+    bool WaitFor(const std::chrono::duration<Duration>& time) {
+        std::unique_lock lk{mutex};
+        if (!condvar.wait_for(lk, time, [this] { return is_set; }))
+            return false;
+        is_set = false;
+        return true;
+    }
+
     template <class Clock, class Duration>
     bool WaitUntil(const std::chrono::time_point<Clock, Duration>& time) {
         std::unique_lock lk{mutex};

src/common/x64/cpu_detect.cpp

@@ -3,8 +3,6 @@
 // Refer to the license.txt file included.
 
 #include <cstring>
-#include <string>
-#include <thread>
 #include "common/common_types.h"
 #include "common/x64/cpu_detect.h"
 
@@ -51,8 +49,6 @@ namespace Common {
 static CPUCaps Detect() {
     CPUCaps caps = {};
 
-    caps.num_cores = std::thread::hardware_concurrency();
-
     // Assumes the CPU supports the CPUID instruction. Those that don't would likely not support
     // yuzu at all anyway
 
@@ -70,12 +66,6 @@ static CPUCaps Detect() {
     __cpuid(cpu_id, 0x80000000);
 
     u32 max_ex_fn = cpu_id[0];
-    if (!strcmp(caps.brand_string, "GenuineIntel"))
-        caps.vendor = CPUVendor::INTEL;
-    else if (!strcmp(caps.brand_string, "AuthenticAMD"))
-        caps.vendor = CPUVendor::AMD;
-    else
-        caps.vendor = CPUVendor::OTHER;
 
     // Set reasonable default brand string even if brand string not available
     strcpy(caps.cpu_string, caps.brand_string);
@@ -96,15 +86,9 @@ static CPUCaps Detect() {
             caps.sse4_1 = true;
         if ((cpu_id[2] >> 20) & 1)
             caps.sse4_2 = true;
-        if ((cpu_id[2] >> 22) & 1)
-            caps.movbe = true;
         if ((cpu_id[2] >> 25) & 1)
             caps.aes = true;
 
-        if ((cpu_id[3] >> 24) & 1) {
-            caps.fxsave_fxrstor = true;
-        }
-
         // AVX support requires 3 separate checks:
         //  - Is the AVX bit set in CPUID?
         //  - Is the XSAVE bit set in CPUID?
@@ -129,8 +113,6 @@ static CPUCaps Detect() {
         }
     }
 
-    caps.flush_to_zero = caps.sse;
-
     if (max_ex_fn >= 0x80000004) {
         // Extract CPU model string
         __cpuid(cpu_id, 0x80000002);
@@ -144,14 +126,8 @@ static CPUCaps Detect() {
     if (max_ex_fn >= 0x80000001) {
         // Check for more features
         __cpuid(cpu_id, 0x80000001);
-        if (cpu_id[2] & 1)
-            caps.lahf_sahf_64 = true;
-        if ((cpu_id[2] >> 5) & 1)
-            caps.lzcnt = true;
         if ((cpu_id[2] >> 16) & 1)
             caps.fma4 = true;
-        if ((cpu_id[3] >> 29) & 1)
-            caps.long_mode = true;
     }
 
     return caps;
@@ -162,48 +138,4 @@ const CPUCaps& GetCPUCaps() {
     return caps;
 }
 
-std::string GetCPUCapsString() {
-    auto caps = GetCPUCaps();
-
-    std::string sum(caps.cpu_string);
-    sum += " (";
-    sum += caps.brand_string;
-    sum += ")";
-
-    if (caps.sse)
-        sum += ", SSE";
-    if (caps.sse2) {
-        sum += ", SSE2";
-        if (!caps.flush_to_zero)
-            sum += " (without DAZ)";
-    }
-
-    if (caps.sse3)
-        sum += ", SSE3";
-    if (caps.ssse3)
-        sum += ", SSSE3";
-    if (caps.sse4_1)
-        sum += ", SSE4.1";
-    if (caps.sse4_2)
-        sum += ", SSE4.2";
-    if (caps.avx)
-        sum += ", AVX";
-    if (caps.avx2)
-        sum += ", AVX2";
-    if (caps.bmi1)
-        sum += ", BMI1";
-    if (caps.bmi2)
-        sum += ", BMI2";
-    if (caps.fma)
-        sum += ", FMA";
-    if (caps.aes)
-        sum += ", AES";
-    if (caps.movbe)
-        sum += ", MOVBE";
-    if (caps.long_mode)
-        sum += ", 64-bit support";
-
-    return sum;
-}
-
 } // namespace Common

src/common/x64/cpu_detect.h

@@ -4,23 +4,12 @@
 
 #pragma once
 
-#include <string>
-
 namespace Common {
 
-/// x86/x64 CPU vendors that may be detected by this module
-enum class CPUVendor {
-    INTEL,
-    AMD,
-    OTHER,
-};
-
 /// x86/x64 CPU capabilities that may be detected by this module
 struct CPUCaps {
-    CPUVendor vendor;
     char cpu_string[0x21];
     char brand_string[0x41];
-    int num_cores;
     bool sse;
     bool sse2;
     bool sse3;
@@ -35,20 +24,6 @@ struct CPUCaps {
     bool fma;
     bool fma4;
     bool aes;
-
-    // Support for the FXSAVE and FXRSTOR instructions
-    bool fxsave_fxrstor;
-
-    bool movbe;
-
-    // This flag indicates that the hardware supports some mode in which denormal inputs and outputs
-    // are automatically set to (signed) zero.
-    bool flush_to_zero;
-
-    // Support for LAHF and SAHF instructions in 64-bit mode
-    bool lahf_sahf_64;
-
-    bool long_mode;
 };
 
 /**
@@ -57,10 +32,4 @@ struct CPUCaps {
  */
 const CPUCaps& GetCPUCaps();
 
-/**
- * Gets a string summary of the name and supported capabilities of the host CPU
- * @return String summary
- */
-std::string GetCPUCapsString();
-
 } // namespace Common

src/core/arm/dynarmic/arm_dynarmic.cpp

@@ -141,6 +141,7 @@ std::unique_ptr<Dynarmic::A64::Jit> ARM_Dynarmic::MakeJit(Common::PageTable& pag
     config.page_table = reinterpret_cast<void**>(page_table.pointers.data());
     config.page_table_address_space_bits = address_space_bits;
     config.silently_mirror_page_table = false;
+    config.absolute_offset_page_table = true;
 
     // Multi-process state
     config.processor_id = core_index;

src/core/frontend/applets/software_keyboard.cpp

@@ -10,8 +10,7 @@ namespace Core::Frontend {
 SoftwareKeyboardApplet::~SoftwareKeyboardApplet() = default;
 
 void DefaultSoftwareKeyboardApplet::RequestText(
-    std::function<void(std::optional<std::u16string>)> out,
-    SoftwareKeyboardParameters parameters) const {
+    std::function<void(std::optional<std::u16string>)> out, SoftwareKeyboardParameters parameters) {
     if (parameters.initial_text.empty())
         out(u"yuzu");
 

src/core/frontend/applets/software_keyboard.h

@@ -38,7 +38,7 @@ public:
     virtual ~SoftwareKeyboardApplet();
 
     virtual void RequestText(std::function<void(std::optional<std::u16string>)> out,
-                             SoftwareKeyboardParameters parameters) const = 0;
+                             SoftwareKeyboardParameters parameters) = 0;
     virtual void SendTextCheckDialog(std::u16string error_message,
                                      std::function<void()> finished_check) const = 0;
 };
@@ -46,7 +46,7 @@ public:
 class DefaultSoftwareKeyboardApplet final : public SoftwareKeyboardApplet {
 public:
     void RequestText(std::function<void(std::optional<std::u16string>)> out,
-                     SoftwareKeyboardParameters parameters) const override;
+                     SoftwareKeyboardParameters parameters) override;
     void SendTextCheckDialog(std::u16string error_message,
                              std::function<void()> finished_check) const override;
 };

src/core/hle/kernel/physical_memory.h

@@ -14,6 +14,9 @@ namespace Kernel {
 // - Second to ensure all host backing memory used is aligned to 256 bytes due
 // to strict alignment restrictions on GPU memory.
 
-using PhysicalMemory = std::vector<u8, Common::AlignmentAllocator<u8, 256>>;
+using PhysicalMemoryVector = std::vector<u8, Common::AlignmentAllocator<u8, 256>>;
+class PhysicalMemory final : public PhysicalMemoryVector {
+    using PhysicalMemoryVector::PhysicalMemoryVector;
+};
 
 } // namespace Kernel

src/core/hle/kernel/process.cpp

@@ -317,6 +317,8 @@ void Process::FreeTLSRegion(VAddr tls_address) {
 }
 
 void Process::LoadModule(CodeSet module_, VAddr base_addr) {
+    code_memory_size += module_.memory.size();
+
     const auto memory = std::make_shared<PhysicalMemory>(std::move(module_.memory));
 
     const auto MapSegment = [&](const CodeSet::Segment& segment, VMAPermission permissions,
@@ -332,8 +334,6 @@ void Process::LoadModule(CodeSet module_, VAddr base_addr) {
     MapSegment(module_.CodeSegment(), VMAPermission::ReadExecute, MemoryState::Code);
     MapSegment(module_.RODataSegment(), VMAPermission::Read, MemoryState::CodeData);
     MapSegment(module_.DataSegment(), VMAPermission::ReadWrite, MemoryState::CodeData);
-
-    code_memory_size += module_.memory.size();
 }
 
 Process::Process(Core::System& system)

src/core/hle/kernel/vm_manager.cpp

@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
+#include <cstring>
 #include <iterator>
 #include <utility>
 #include "common/alignment.h"
@@ -269,18 +270,9 @@ ResultVal<VAddr> VMManager::SetHeapSize(u64 size) {
     // If necessary, expand backing vector to cover new heap extents in
     // the case of allocating. Otherwise, shrink the backing memory,
     // if a smaller heap has been requested.
-    const u64 old_heap_size = GetCurrentHeapSize();
-    if (size > old_heap_size) {
-        const u64 alloc_size = size - old_heap_size;
-
-        heap_memory->insert(heap_memory->end(), alloc_size, 0);
-        RefreshMemoryBlockMappings(heap_memory.get());
-    } else if (size < old_heap_size) {
-        heap_memory->resize(size);
-        heap_memory->shrink_to_fit();
-
-        RefreshMemoryBlockMappings(heap_memory.get());
-    }
+    heap_memory->resize(size);
+    heap_memory->shrink_to_fit();
+    RefreshMemoryBlockMappings(heap_memory.get());
 
     heap_end = heap_region_base + size;
     ASSERT(GetCurrentHeapSize() == heap_memory->size());
@@ -752,24 +744,20 @@ void VMManager::MergeAdjacentVMA(VirtualMemoryArea& left, const VirtualMemoryAre
     // Always merge allocated memory blocks, even when they don't share the same backing block.
     if (left.type == VMAType::AllocatedMemoryBlock &&
         (left.backing_block != right.backing_block || left.offset + left.size != right.offset)) {
-        const auto right_begin = right.backing_block->begin() + right.offset;
-        const auto right_end = right_begin + right.size;
 
         // Check if we can save work.
         if (left.offset == 0 && left.size == left.backing_block->size()) {
             // Fast case: left is an entire backing block.
-            left.backing_block->insert(left.backing_block->end(), right_begin, right_end);
+            left.backing_block->resize(left.size + right.size);
+            std::memcpy(left.backing_block->data() + left.size,
+                        right.backing_block->data() + right.offset, right.size);
         } else {
             // Slow case: make a new memory block for left and right.
-            const auto left_begin = left.backing_block->begin() + left.offset;
-            const auto left_end = left_begin + left.size;
-            const auto left_size = static_cast<std::size_t>(std::distance(left_begin, left_end));
-            const auto right_size = static_cast<std::size_t>(std::distance(right_begin, right_end));
-
             auto new_memory = std::make_shared<PhysicalMemory>();
-            new_memory->reserve(left_size + right_size);
-            new_memory->insert(new_memory->end(), left_begin, left_end);
-            new_memory->insert(new_memory->end(), right_begin, right_end);
+            new_memory->resize(left.size + right.size);
+            std::memcpy(new_memory->data(), left.backing_block->data() + left.offset, left.size);
+            std::memcpy(new_memory->data() + left.size, right.backing_block->data() + right.offset,
+                        right.size);
 
             left.backing_block = std::move(new_memory);
             left.offset = 0;
@@ -792,8 +780,7 @@ void VMManager::UpdatePageTableForVMA(const VirtualMemoryArea& vma) {
         memory.UnmapRegion(page_table, vma.base, vma.size);
         break;
     case VMAType::AllocatedMemoryBlock:
-        memory.MapMemoryRegion(page_table, vma.base, vma.size,
-                               vma.backing_block->data() + vma.offset);
+        memory.MapMemoryRegion(page_table, vma.base, vma.size, *vma.backing_block, vma.offset);
         break;
     case VMAType::BackingMemory:
         memory.MapMemoryRegion(page_table, vma.base, vma.size, vma.backing_memory);

src/core/hle/service/am/am.cpp

@@ -778,11 +778,9 @@ private:
     void GetAppletStateChangedEvent(Kernel::HLERequestContext& ctx) {
         LOG_DEBUG(Service_AM, "called");
 
-        const auto event = applet->GetBroker().GetStateChangedEvent();
-
         IPC::ResponseBuilder rb{ctx, 2, 1};
         rb.Push(RESULT_SUCCESS);
-        rb.PushCopyObjects(event);
+        rb.PushCopyObjects(applet->GetBroker().GetStateChangedEvent()->GetReadableEvent());
     }
 
     void IsCompleted(Kernel::HLERequestContext& ctx) {
@@ -836,6 +834,8 @@ private:
             return;
         }
 
+        applet->GetBroker().GetNormalDataEvent()->Clear();
+
         rb.Push(RESULT_SUCCESS);
         rb.PushIpcInterface<IStorage>(std::move(*storage));
     }
@@ -868,6 +868,8 @@ private:
             return;
         }
 
+        applet->GetBroker().GetInteractiveDataEvent()->Clear();
+
         rb.Push(RESULT_SUCCESS);
         rb.PushIpcInterface<IStorage>(std::move(*storage));
     }
@@ -877,7 +879,7 @@ private:
 
         IPC::ResponseBuilder rb{ctx, 2, 1};
         rb.Push(RESULT_SUCCESS);
-        rb.PushCopyObjects(applet->GetBroker().GetNormalDataEvent());
+        rb.PushCopyObjects(applet->GetBroker().GetNormalDataEvent()->GetReadableEvent());
     }
 
     void GetPopInteractiveOutDataEvent(Kernel::HLERequestContext& ctx) {
@@ -885,7 +887,7 @@ private:
 
         IPC::ResponseBuilder rb{ctx, 2, 1};
         rb.Push(RESULT_SUCCESS);
-        rb.PushCopyObjects(applet->GetBroker().GetInteractiveDataEvent());
+        rb.PushCopyObjects(applet->GetBroker().GetInteractiveDataEvent()->GetReadableEvent());
     }
 
     std::shared_ptr<Applets::Applet> applet;

src/core/hle/service/am/applets/applets.cpp

@@ -108,16 +108,16 @@ void AppletDataBroker::SignalStateChanged() const {
     state_changed_event.writable->Signal();
 }
 
-std::shared_ptr<Kernel::ReadableEvent> AppletDataBroker::GetNormalDataEvent() const {
-    return pop_out_data_event.readable;
+std::shared_ptr<Kernel::WritableEvent> AppletDataBroker::GetNormalDataEvent() const {
+    return pop_out_data_event.writable;
 }
 
-std::shared_ptr<Kernel::ReadableEvent> AppletDataBroker::GetInteractiveDataEvent() const {
-    return pop_interactive_out_data_event.readable;
+std::shared_ptr<Kernel::WritableEvent> AppletDataBroker::GetInteractiveDataEvent() const {
+    return pop_interactive_out_data_event.writable;
 }
 
-std::shared_ptr<Kernel::ReadableEvent> AppletDataBroker::GetStateChangedEvent() const {
-    return state_changed_event.readable;
+std::shared_ptr<Kernel::WritableEvent> AppletDataBroker::GetStateChangedEvent() const {
+    return state_changed_event.writable;
 }
 
 Applet::Applet(Kernel::KernelCore& kernel_) : broker{kernel_} {}

src/core/hle/service/am/applets/applets.h

@@ -86,9 +86,9 @@ public:
 
     void SignalStateChanged() const;
 
-    std::shared_ptr<Kernel::ReadableEvent> GetNormalDataEvent() const;
-    std::shared_ptr<Kernel::ReadableEvent> GetInteractiveDataEvent() const;
-    std::shared_ptr<Kernel::ReadableEvent> GetStateChangedEvent() const;
+    std::shared_ptr<Kernel::WritableEvent> GetNormalDataEvent() const;
+    std::shared_ptr<Kernel::WritableEvent> GetInteractiveDataEvent() const;
+    std::shared_ptr<Kernel::WritableEvent> GetStateChangedEvent() const;
 
 private:
     // Queues are named from applet's perspective

src/core/hle/service/am/applets/software_keyboard.cpp

@@ -40,8 +40,8 @@ static Core::Frontend::SoftwareKeyboardParameters ConvertToFrontendParameters(
 }
 
 SoftwareKeyboard::SoftwareKeyboard(Core::System& system_,
-                                   const Core::Frontend::SoftwareKeyboardApplet& frontend_)
-    : Applet{system_.Kernel()}, frontend(frontend_) {}
+                                   Core::Frontend::SoftwareKeyboardApplet& frontend)
+    : Applet{system_.Kernel()}, frontend{frontend} {}
 
 SoftwareKeyboard::~SoftwareKeyboard() = default;
 
@@ -103,6 +103,7 @@ void SoftwareKeyboard::ExecuteInteractive() {
 void SoftwareKeyboard::Execute() {
     if (complete) {
         broker.PushNormalDataFromApplet(IStorage{final_data});
+        broker.SignalStateChanged();
         return;
     }
 
@@ -119,7 +120,7 @@ void SoftwareKeyboard::WriteText(std::optional<std::u16string> text) {
         std::vector<u8> output_sub(SWKBD_OUTPUT_BUFFER_SIZE);
 
         if (config.utf_8) {
-            const u64 size = text->size() + 8;
+            const u64 size = text->size() + sizeof(u64);
             const auto new_text = Common::UTF16ToUTF8(*text);
 
             std::memcpy(output_sub.data(), &size, sizeof(u64));
@@ -130,7 +131,7 @@ void SoftwareKeyboard::WriteText(std::optional<std::u16string> text) {
             std::memcpy(output_main.data() + 4, new_text.data(),
                         std::min(new_text.size(), SWKBD_OUTPUT_BUFFER_SIZE - 4));
         } else {
-            const u64 size = text->size() * 2 + 8;
+            const u64 size = text->size() * 2 + sizeof(u64);
             std::memcpy(output_sub.data(), &size, sizeof(u64));
             std::memcpy(output_sub.data() + 8, text->data(),
                         std::min(text->size() * 2, SWKBD_OUTPUT_BUFFER_SIZE - 8));

src/core/hle/service/am/applets/software_keyboard.h

@@ -60,7 +60,7 @@ static_assert(sizeof(KeyboardConfig) == 0x3E0, "KeyboardConfig has incorrect siz
 class SoftwareKeyboard final : public Applet {
 public:
     explicit SoftwareKeyboard(Core::System& system_,
-                              const Core::Frontend::SoftwareKeyboardApplet& frontend_);
+                              Core::Frontend::SoftwareKeyboardApplet& frontend);
     ~SoftwareKeyboard() override;
 
     void Initialize() override;
@@ -73,7 +73,7 @@ public:
     void WriteText(std::optional<std::u16string> text);
 
 private:
-    const Core::Frontend::SoftwareKeyboardApplet& frontend;
+    Core::Frontend::SoftwareKeyboardApplet& frontend;
 
     KeyboardConfig config;
     std::u16string initial_text;

src/core/hle/service/nifm/nifm.cpp

@@ -9,6 +9,7 @@
 #include "core/hle/kernel/writable_event.h"
 #include "core/hle/service/nifm/nifm.h"
 #include "core/hle/service/service.h"
+#include "core/settings.h"
 
 namespace Service::NIFM {
 
@@ -86,7 +87,12 @@ private:
 
         IPC::ResponseBuilder rb{ctx, 3};
         rb.Push(RESULT_SUCCESS);
-        rb.PushEnum(RequestState::Connected);
+
+        if (Settings::values.bcat_backend == "none") {
+            rb.PushEnum(RequestState::NotSubmitted);
+        } else {
+            rb.PushEnum(RequestState::Connected);
+        }
     }
 
     void GetResult(Kernel::HLERequestContext& ctx) {
@@ -194,14 +200,22 @@ private:
 
         IPC::ResponseBuilder rb{ctx, 3};
         rb.Push(RESULT_SUCCESS);
-        rb.Push<u8>(1);
+        if (Settings::values.bcat_backend == "none") {
+            rb.Push<u8>(0);
+        } else {
+            rb.Push<u8>(1);
+        }
     }
     void IsAnyInternetRequestAccepted(Kernel::HLERequestContext& ctx) {
         LOG_WARNING(Service_NIFM, "(STUBBED) called");
 
         IPC::ResponseBuilder rb{ctx, 3};
         rb.Push(RESULT_SUCCESS);
-        rb.Push<u8>(1);
+        if (Settings::values.bcat_backend == "none") {
+            rb.Push<u8>(0);
+        } else {
+            rb.Push<u8>(1);
+        }
     }
     Core::System& system;
 };

src/core/hle/service/nvflinger/nvflinger.cpp

@@ -88,6 +88,12 @@ std::optional<u64> NVFlinger::CreateLayer(u64 display_id) {
     return layer_id;
 }
 
+void NVFlinger::CloseLayer(u64 layer_id) {
+    for (auto& display : displays) {
+        display.CloseLayer(layer_id);
+    }
+}
+
 std::optional<u32> NVFlinger::FindBufferQueueId(u64 display_id, u64 layer_id) const {
     const auto* const layer = FindLayer(display_id, layer_id);
 
@@ -192,7 +198,7 @@ void NVFlinger::Compose() {
 
         const auto& igbp_buffer = buffer->get().igbp_buffer;
 
-        const auto& gpu = system.GPU();
+        auto& gpu = system.GPU();
         const auto& multi_fence = buffer->get().multi_fence;
         for (u32 fence_id = 0; fence_id < multi_fence.num_fences; fence_id++) {
             const auto& fence = multi_fence.fences[fence_id];

src/core/hle/service/nvflinger/nvflinger.h

@@ -54,6 +54,9 @@ public:
     /// If an invalid display ID is specified, then an empty optional is returned.
     std::optional<u64> CreateLayer(u64 display_id);
 
+    /// Closes a layer on all displays for the given layer ID.
+    void CloseLayer(u64 layer_id);
+
     /// Finds the buffer queue ID of the specified layer in the specified display.
     ///
     /// If an invalid display ID or layer ID is provided, then an empty optional is returned.

src/core/hle/service/time/time_zone_manager.cpp

@@ -820,7 +820,7 @@ static ResultCode ToCalendarTimeImpl(const TimeZoneRule& rules, s64 time, Calend
     const ResultCode result{
         ToCalendarTimeInternal(rules, time, calendar_time, calendar.additiona_info)};
     calendar.time.year = static_cast<s16>(calendar_time.year);
-    calendar.time.month = calendar_time.month;
+    calendar.time.month = calendar_time.month + 1; // Internal impl. uses 0-indexed month
     calendar.time.day = calendar_time.day;
     calendar.time.hour = calendar_time.hour;
     calendar.time.minute = calendar_time.minute;
@@ -874,7 +874,7 @@ ResultCode TimeZoneManager::ToPosixTime(const TimeZoneRule& rules,
 
     CalendarTimeInternal internal_time{};
     internal_time.year = calendar_time.year;
-    internal_time.month = calendar_time.month;
+    internal_time.month = calendar_time.month - 1; // Internal impl. uses 0-indexed month
     internal_time.day = calendar_time.day;
     internal_time.hour = calendar_time.hour;
     internal_time.minute = calendar_time.minute;
@@ -1019,6 +1019,15 @@ ResultCode TimeZoneManager::ToPosixTime(const TimeZoneRule& rules,
     return RESULT_SUCCESS;
 }
 
+ResultCode TimeZoneManager::ToPosixTimeWithMyRule(const CalendarTime& calendar_time,
+                                                  s64& posix_time) const {
+    if (is_initialized) {
+        return ToPosixTime(time_zone_rule, calendar_time, posix_time);
+    }
+    posix_time = 0;
+    return ERROR_UNINITIALIZED_CLOCK;
+}
+
 ResultCode TimeZoneManager::GetDeviceLocationName(LocationName& value) const {
     if (!is_initialized) {
         return ERROR_UNINITIALIZED_CLOCK;

src/core/hle/service/time/time_zone_manager.h

@@ -39,6 +39,7 @@ public:
     ResultCode ParseTimeZoneRuleBinary(TimeZoneRule& rules, FileSys::VirtualFile& vfs_file) const;
     ResultCode ToPosixTime(const TimeZoneRule& rules, const CalendarTime& calendar_time,
                            s64& posix_time) const;
+    ResultCode ToPosixTimeWithMyRule(const CalendarTime& calendar_time, s64& posix_time) const;
 
 private:
     bool is_initialized{};

src/core/hle/service/time/time_zone_service.cpp

@@ -22,7 +22,7 @@ ITimeZoneService ::ITimeZoneService(TimeZone::TimeZoneContentManager& time_zone_
         {100, &ITimeZoneService::ToCalendarTime, "ToCalendarTime"},
         {101, &ITimeZoneService::ToCalendarTimeWithMyRule, "ToCalendarTimeWithMyRule"},
         {201, &ITimeZoneService::ToPosixTime, "ToPosixTime"},
-        {202, nullptr, "ToPosixTimeWithMyRule"},
+        {202, &ITimeZoneService::ToPosixTimeWithMyRule, "ToPosixTimeWithMyRule"},
     };
     RegisterHandlers(functions);
 }
@@ -145,4 +145,26 @@ void ITimeZoneService::ToPosixTime(Kernel::HLERequestContext& ctx) {
     ctx.WriteBuffer(&posix_time, sizeof(s64));
 }
 
+void ITimeZoneService::ToPosixTimeWithMyRule(Kernel::HLERequestContext& ctx) {
+    LOG_DEBUG(Service_Time, "called");
+
+    IPC::RequestParser rp{ctx};
+    const auto calendar_time{rp.PopRaw<TimeZone::CalendarTime>()};
+
+    s64 posix_time{};
+    if (const ResultCode result{
+            time_zone_content_manager.GetTimeZoneManager().ToPosixTimeWithMyRule(calendar_time,
+                                                                                 posix_time)};
+        result != RESULT_SUCCESS) {
+        IPC::ResponseBuilder rb{ctx, 2};
+        rb.Push(result);
+        return;
+    }
+
+    IPC::ResponseBuilder rb{ctx, 3};
+    rb.Push(RESULT_SUCCESS);
+    rb.PushRaw<u32>(1); // Number of times we're returning
+    ctx.WriteBuffer(&posix_time, sizeof(s64));
+}
+
 } // namespace Service::Time

src/core/hle/service/time/time_zone_service.h

@@ -22,6 +22,7 @@ private:
     void ToCalendarTime(Kernel::HLERequestContext& ctx);
     void ToCalendarTimeWithMyRule(Kernel::HLERequestContext& ctx);
     void ToPosixTime(Kernel::HLERequestContext& ctx);
+    void ToPosixTimeWithMyRule(Kernel::HLERequestContext& ctx);
 
 private:
     TimeZone::TimeZoneContentManager& time_zone_content_manager;

src/core/hle/service/vi/display/vi_display.cpp

@@ -24,11 +24,11 @@ Display::Display(u64 id, std::string name, Core::System& system) : id{id}, name{
 Display::~Display() = default;
 
 Layer& Display::GetLayer(std::size_t index) {
-    return layers.at(index);
+    return *layers.at(index);
 }
 
 const Layer& Display::GetLayer(std::size_t index) const {
-    return layers.at(index);
+    return *layers.at(index);
 }
 
 std::shared_ptr<Kernel::ReadableEvent> Display::GetVSyncEvent() const {
@@ -43,29 +43,38 @@ void Display::CreateLayer(u64 id, NVFlinger::BufferQueue& buffer_queue) {
     // TODO(Subv): Support more than 1 layer.
     ASSERT_MSG(layers.empty(), "Only one layer is supported per display at the moment");
 
-    layers.emplace_back(id, buffer_queue);
+    layers.emplace_back(std::make_shared<Layer>(id, buffer_queue));
+}
+
+void Display::CloseLayer(u64 id) {
+    layers.erase(
+        std::remove_if(layers.begin(), layers.end(),
+                       [id](const std::shared_ptr<Layer>& layer) { return layer->GetID() == id; }),
+        layers.end());
 }
 
 Layer* Display::FindLayer(u64 id) {
-    const auto itr = std::find_if(layers.begin(), layers.end(),
-                                  [id](const VI::Layer& layer) { return layer.GetID() == id; });
+    const auto itr =
+        std::find_if(layers.begin(), layers.end(),
+                     [id](const std::shared_ptr<Layer>& layer) { return layer->GetID() == id; });
 
     if (itr == layers.end()) {
         return nullptr;
     }
 
-    return &*itr;
+    return itr->get();
 }
 
 const Layer* Display::FindLayer(u64 id) const {
-    const auto itr = std::find_if(layers.begin(), layers.end(),
-                                  [id](const VI::Layer& layer) { return layer.GetID() == id; });
+    const auto itr =
+        std::find_if(layers.begin(), layers.end(),
+                     [id](const std::shared_ptr<Layer>& layer) { return layer->GetID() == id; });
 
     if (itr == layers.end()) {
         return nullptr;
     }
 
-    return &*itr;
+    return itr->get();
 }
 
 } // namespace Service::VI

src/core/hle/service/vi/display/vi_display.h

@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include <memory>
 #include <string>
 #include <vector>
 
@@ -69,6 +70,12 @@ public:
     ///
     void CreateLayer(u64 id, NVFlinger::BufferQueue& buffer_queue);
 
+    /// Closes and removes a layer from this display with the given ID.
+    ///
+    /// @param id           The ID assigned to the layer to close.
+    ///
+    void CloseLayer(u64 id);
+
     /// Attempts to find a layer with the given ID.
     ///
     /// @param id The layer ID.
@@ -91,7 +98,7 @@ private:
     u64 id;
     std::string name;
 
-    std::vector<Layer> layers;
+    std::vector<std::shared_ptr<Layer>> layers;
     Kernel::EventPair vsync_event;
 };
 

src/core/hle/service/vi/vi.cpp

@@ -1066,6 +1066,18 @@ private:
         rb.Push<u64>(ctx.WriteBuffer(native_window.Serialize()));
     }
 
+    void CloseLayer(Kernel::HLERequestContext& ctx) {
+        IPC::RequestParser rp{ctx};
+        const auto layer_id{rp.Pop<u64>()};
+
+        LOG_DEBUG(Service_VI, "called. layer_id=0x{:016X}", layer_id);
+
+        nv_flinger->CloseLayer(layer_id);
+
+        IPC::ResponseBuilder rb{ctx, 2};
+        rb.Push(RESULT_SUCCESS);
+    }
+
     void CreateStrayLayer(Kernel::HLERequestContext& ctx) {
         IPC::RequestParser rp{ctx};
         const u32 flags = rp.Pop<u32>();
@@ -1178,7 +1190,7 @@ IApplicationDisplayService::IApplicationDisplayService(
         {1101, &IApplicationDisplayService::SetDisplayEnabled, "SetDisplayEnabled"},
         {1102, &IApplicationDisplayService::GetDisplayResolution, "GetDisplayResolution"},
         {2020, &IApplicationDisplayService::OpenLayer, "OpenLayer"},
-        {2021, nullptr, "CloseLayer"},
+        {2021, &IApplicationDisplayService::CloseLayer, "CloseLayer"},
         {2030, &IApplicationDisplayService::CreateStrayLayer, "CreateStrayLayer"},
         {2031, &IApplicationDisplayService::DestroyStrayLayer, "DestroyStrayLayer"},
         {2101, &IApplicationDisplayService::SetLayerScalingMode, "SetLayerScalingMode"},

src/core/loader/elf.cpp

@@ -335,7 +335,8 @@ Kernel::CodeSet ElfReader::LoadInto(VAddr vaddr) {
             codeset_segment->addr = segment_addr;
             codeset_segment->size = aligned_size;
 
-            memcpy(&program_image[current_image_position], GetSegmentPtr(i), p->p_filesz);
+            std::memcpy(program_image.data() + current_image_position, GetSegmentPtr(i),
+                        p->p_filesz);
             current_image_position += aligned_size;
         }
     }

src/core/loader/kip.cpp

@@ -2,6 +2,7 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <cstring>
 #include "core/file_sys/kernel_executable.h"
 #include "core/file_sys/program_metadata.h"
 #include "core/gdbstub/gdbstub.h"
@@ -76,8 +77,8 @@ AppLoader::LoadResult AppLoader_KIP::Load(Kernel::Process& process) {
         segment.addr = offset;
         segment.offset = offset;
         segment.size = PageAlignSize(static_cast<u32>(data.size()));
-        program_image.resize(offset);
-        program_image.insert(program_image.end(), data.begin(), data.end());
+        program_image.resize(offset + data.size());
+        std::memcpy(program_image.data() + offset, data.data(), data.size());
     };
 
     load_segment(codeset.CodeSegment(), kip->GetTextSection(), kip->GetTextOffset());

src/core/loader/nso.cpp

@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.
 
 #include <cinttypes>
+#include <cstring>
 #include <vector>
 
 #include "common/common_funcs.h"
@@ -96,15 +97,21 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::Process& process,
         if (nso_header.IsSegmentCompressed(i)) {
             data = DecompressSegment(data, nso_header.segments[i]);
         }
-        program_image.resize(nso_header.segments[i].location);
-        program_image.insert(program_image.end(), data.begin(), data.end());
+        program_image.resize(nso_header.segments[i].location +
+                             PageAlignSize(static_cast<u32>(data.size())));
+        std::memcpy(program_image.data() + nso_header.segments[i].location, data.data(),
+                    data.size());
         codeset.segments[i].addr = nso_header.segments[i].location;
         codeset.segments[i].offset = nso_header.segments[i].location;
         codeset.segments[i].size = PageAlignSize(static_cast<u32>(data.size()));
     }
 
-    if (should_pass_arguments && !Settings::values.program_args.empty()) {
-        const auto arg_data = Settings::values.program_args;
+    if (should_pass_arguments) {
+        std::vector<u8> arg_data{Settings::values.program_args.begin(),
+                                 Settings::values.program_args.end()};
+        if (arg_data.empty()) {
+            arg_data.resize(NSO_ARGUMENT_DEFAULT_SIZE);
+        }
         codeset.DataSegment().size += NSO_ARGUMENT_DATA_ALLOCATION_SIZE;
         NSOArgumentHeader args_header{
             NSO_ARGUMENT_DATA_ALLOCATION_SIZE, static_cast<u32_le>(arg_data.size()), {}};
@@ -139,12 +146,12 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::Process& process,
         std::vector<u8> pi_header;
         pi_header.insert(pi_header.begin(), reinterpret_cast<u8*>(&nso_header),
                          reinterpret_cast<u8*>(&nso_header) + sizeof(NSOHeader));
-        pi_header.insert(pi_header.begin() + sizeof(NSOHeader), program_image.begin(),
-                         program_image.end());
+        pi_header.insert(pi_header.begin() + sizeof(NSOHeader), program_image.data(),
+                         program_image.data() + program_image.size());
 
         pi_header = pm->PatchNSO(pi_header, file.GetName());
 
-        std::copy(pi_header.begin() + sizeof(NSOHeader), pi_header.end(), program_image.begin());
+        std::copy(pi_header.begin() + sizeof(NSOHeader), pi_header.end(), program_image.data());
     }
 
     // Apply cheats if they exist and the program has a valid title ID

src/core/loader/nso.h

@@ -56,6 +56,8 @@ static_assert(sizeof(NSOHeader) == 0x100, "NSOHeader has incorrect size.");
 static_assert(std::is_trivially_copyable_v<NSOHeader>, "NSOHeader must be trivially copyable.");
 
 constexpr u64 NSO_ARGUMENT_DATA_ALLOCATION_SIZE = 0x9000;
+// NOTE: Official software default argument state is unverified.
+constexpr u64 NSO_ARGUMENT_DEFAULT_SIZE = 1;
 
 struct NSOArgumentHeader {
     u32_le allocated_size;

src/core/memory.cpp

@@ -14,6 +14,7 @@
 #include "common/swap.h"
 #include "core/arm/arm_interface.h"
 #include "core/core.h"
+#include "core/hle/kernel/physical_memory.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/memory.h"
@@ -38,6 +39,11 @@ struct Memory::Impl {
         system.ArmInterface(3).PageTableChanged(*current_page_table, address_space_width);
     }
 
+    void MapMemoryRegion(Common::PageTable& page_table, VAddr base, u64 size,
+                         Kernel::PhysicalMemory& memory, VAddr offset) {
+        MapMemoryRegion(page_table, base, size, memory.data() + offset);
+    }
+
     void MapMemoryRegion(Common::PageTable& page_table, VAddr base, u64 size, u8* target) {
         ASSERT_MSG((size & PAGE_MASK) == 0, "non-page aligned size: {:016X}", size);
         ASSERT_MSG((base & PAGE_MASK) == 0, "non-page aligned base: {:016X}", base);
@@ -146,7 +152,7 @@ struct Memory::Impl {
     u8* GetPointer(const VAddr vaddr) {
         u8* const page_pointer = current_page_table->pointers[vaddr >> PAGE_BITS];
         if (page_pointer != nullptr) {
-            return page_pointer + (vaddr & PAGE_MASK);
+            return page_pointer + vaddr;
         }
 
         if (current_page_table->attributes[vaddr >> PAGE_BITS] ==
@@ -229,7 +235,8 @@ struct Memory::Impl {
             case Common::PageType::Memory: {
                 DEBUG_ASSERT(page_table.pointers[page_index]);
 
-                const u8* const src_ptr = page_table.pointers[page_index] + page_offset;
+                const u8* const src_ptr =
+                    page_table.pointers[page_index] + page_offset + (page_index << PAGE_BITS);
                 std::memcpy(dest_buffer, src_ptr, copy_amount);
                 break;
             }
@@ -276,7 +283,8 @@ struct Memory::Impl {
             case Common::PageType::Memory: {
                 DEBUG_ASSERT(page_table.pointers[page_index]);
 
-                u8* const dest_ptr = page_table.pointers[page_index] + page_offset;
+                u8* const dest_ptr =
+                    page_table.pointers[page_index] + page_offset + (page_index << PAGE_BITS);
                 std::memcpy(dest_ptr, src_buffer, copy_amount);
                 break;
             }
@@ -322,7 +330,8 @@ struct Memory::Impl {
             case Common::PageType::Memory: {
                 DEBUG_ASSERT(page_table.pointers[page_index]);
 
-                u8* dest_ptr = page_table.pointers[page_index] + page_offset;
+                u8* dest_ptr =
+                    page_table.pointers[page_index] + page_offset + (page_index << PAGE_BITS);
                 std::memset(dest_ptr, 0, copy_amount);
                 break;
             }
@@ -368,7 +377,8 @@ struct Memory::Impl {
             }
             case Common::PageType::Memory: {
                 DEBUG_ASSERT(page_table.pointers[page_index]);
-                const u8* src_ptr = page_table.pointers[page_index] + page_offset;
+                const u8* src_ptr =
+                    page_table.pointers[page_index] + page_offset + (page_index << PAGE_BITS);
                 WriteBlock(process, dest_addr, src_ptr, copy_amount);
                 break;
             }
@@ -446,7 +456,8 @@ struct Memory::Impl {
                         page_type = Common::PageType::Unmapped;
                     } else {
                         page_type = Common::PageType::Memory;
-                        current_page_table->pointers[vaddr >> PAGE_BITS] = pointer;
+                        current_page_table->pointers[vaddr >> PAGE_BITS] =
+                            pointer - (vaddr & ~PAGE_MASK);
                     }
                     break;
                 }
@@ -493,7 +504,9 @@ struct Memory::Impl {
                       memory);
         } else {
             while (base != end) {
-                page_table.pointers[base] = memory;
+                page_table.pointers[base] = memory - (base << PAGE_BITS);
+                ASSERT_MSG(page_table.pointers[base],
+                           "memory mapping base yield a nullptr within the table");
 
                 base += 1;
                 memory += PAGE_SIZE;
@@ -518,7 +531,7 @@ struct Memory::Impl {
         if (page_pointer != nullptr) {
             // NOTE: Avoid adding any extra logic to this fast-path block
             T value;
-            std::memcpy(&value, &page_pointer[vaddr & PAGE_MASK], sizeof(T));
+            std::memcpy(&value, &page_pointer[vaddr], sizeof(T));
             return value;
         }
 
@@ -559,7 +572,7 @@ struct Memory::Impl {
         u8* const page_pointer = current_page_table->pointers[vaddr >> PAGE_BITS];
         if (page_pointer != nullptr) {
             // NOTE: Avoid adding any extra logic to this fast-path block
-            std::memcpy(&page_pointer[vaddr & PAGE_MASK], &data, sizeof(T));
+            std::memcpy(&page_pointer[vaddr], &data, sizeof(T));
             return;
         }
 
@@ -594,6 +607,11 @@ void Memory::SetCurrentPageTable(Kernel::Process& process) {
     impl->SetCurrentPageTable(process);
 }
 
+void Memory::MapMemoryRegion(Common::PageTable& page_table, VAddr base, u64 size,
+                             Kernel::PhysicalMemory& memory, VAddr offset) {
+    impl->MapMemoryRegion(page_table, base, size, memory, offset);
+}
+
 void Memory::MapMemoryRegion(Common::PageTable& page_table, VAddr base, u64 size, u8* target) {
     impl->MapMemoryRegion(page_table, base, size, target);
 }

src/core/memory.h

@@ -19,8 +19,9 @@ class System;
 }
 
 namespace Kernel {
+class PhysicalMemory;
 class Process;
-}
+} // namespace Kernel
 
 namespace Memory {
 
@@ -65,6 +66,19 @@ public:
      */
     void SetCurrentPageTable(Kernel::Process& process);
 
+    /**
+     * Maps an physical buffer onto a region of the emulated process address space.
+     *
+     * @param page_table The page table of the emulated process.
+     * @param base       The address to start mapping at. Must be page-aligned.
+     * @param size       The amount of bytes to map. Must be page-aligned.
+     * @param memory     Physical buffer with the memory backing the mapping. Must be of length
+     *                   at least `size + offset`.
+     * @param offset     The offset within the physical memory. Must be page-aligned.
+     */
+    void MapMemoryRegion(Common::PageTable& page_table, VAddr base, u64 size,
+                         Kernel::PhysicalMemory& memory, VAddr offset);
+
     /**
      * Maps an allocated buffer onto a region of the emulated process address space.
      *

src/core/settings.h

@@ -401,6 +401,9 @@ struct Values {
     std::string motion_device;
     TouchscreenInput touchscreen;
     std::atomic_bool is_device_reload_pending{true};
+    std::string udp_input_address;
+    u16 udp_input_port;
+    u8 udp_pad_index;
 
     // Core
     bool use_multi_core;

src/input_common/CMakeLists.txt

@@ -9,6 +9,12 @@ add_library(input_common STATIC
     motion_emu.h
     sdl/sdl.cpp
     sdl/sdl.h
+    udp/client.cpp
+    udp/client.h
+    udp/protocol.cpp
+    udp/protocol.h
+    udp/udp.cpp
+    udp/udp.h
 )
 
 if(SDL2_FOUND)
@@ -21,4 +27,4 @@ if(SDL2_FOUND)
 endif()
 
 create_target_directory_groups(input_common)
-target_link_libraries(input_common PUBLIC core PRIVATE common)
+target_link_libraries(input_common PUBLIC core PRIVATE common ${Boost_LIBRARIES})

src/input_common/main.cpp

@@ -9,6 +9,7 @@
 #include "input_common/keyboard.h"
 #include "input_common/main.h"
 #include "input_common/motion_emu.h"
+#include "input_common/udp/udp.h"
 #ifdef HAVE_SDL2
 #include "input_common/sdl/sdl.h"
 #endif
@@ -18,6 +19,7 @@ namespace InputCommon {
 static std::shared_ptr<Keyboard> keyboard;
 static std::shared_ptr<MotionEmu> motion_emu;
 static std::unique_ptr<SDL::State> sdl;
+static std::unique_ptr<CemuhookUDP::State> udp;
 
 void Init() {
     keyboard = std::make_shared<Keyboard>();
@@ -28,6 +30,8 @@ void Init() {
     Input::RegisterFactory<Input::MotionDevice>("motion_emu", motion_emu);
 
     sdl = SDL::Init();
+
+    udp = CemuhookUDP::Init();
 }
 
 void Shutdown() {
@@ -72,11 +76,13 @@ std::string GenerateAnalogParamFromKeys(int key_up, int key_down, int key_left,
 namespace Polling {
 
 std::vector<std::unique_ptr<DevicePoller>> GetPollers(DeviceType type) {
+    std::vector<std::unique_ptr<DevicePoller>> pollers;
+
 #ifdef HAVE_SDL2
-    return sdl->GetPollers(type);
-#else
-    return {};
+    pollers = sdl->GetPollers(type);
 #endif
+
+    return pollers;
 }
 
 } // namespace Polling

src/input_common/udp/client.cpp

@@ -0,0 +1,287 @@
+// Copyright 2018 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <array>
+#include <chrono>
+#include <cstring>
+#include <functional>
+#include <thread>
+#include <boost/asio.hpp>
+#include <boost/bind.hpp>
+#include "common/logging/log.h"
+#include "input_common/udp/client.h"
+#include "input_common/udp/protocol.h"
+
+using boost::asio::ip::address_v4;
+using boost::asio::ip::udp;
+
+namespace InputCommon::CemuhookUDP {
+
+struct SocketCallback {
+    std::function<void(Response::Version)> version;
+    std::function<void(Response::PortInfo)> port_info;
+    std::function<void(Response::PadData)> pad_data;
+};
+
+class Socket {
+public:
+    using clock = std::chrono::system_clock;
+
+    explicit Socket(const std::string& host, u16 port, u8 pad_index, u32 client_id,
+                    SocketCallback callback)
+        : client_id(client_id), timer(io_service),
+          send_endpoint(udp::endpoint(address_v4::from_string(host), port)),
+          socket(io_service, udp::endpoint(udp::v4(), 0)), pad_index(pad_index),
+          callback(std::move(callback)) {}
+
+    void Stop() {
+        io_service.stop();
+    }
+
+    void Loop() {
+        io_service.run();
+    }
+
+    void StartSend(const clock::time_point& from) {
+        timer.expires_at(from + std::chrono::seconds(3));
+        timer.async_wait([this](const boost::system::error_code& error) { HandleSend(error); });
+    }
+
+    void StartReceive() {
+        socket.async_receive_from(
+            boost::asio::buffer(receive_buffer), receive_endpoint,
+            [this](const boost::system::error_code& error, std::size_t bytes_transferred) {
+                HandleReceive(error, bytes_transferred);
+            });
+    }
+
+private:
+    void HandleReceive(const boost::system::error_code& error, std::size_t bytes_transferred) {
+        if (auto type = Response::Validate(receive_buffer.data(), bytes_transferred)) {
+            switch (*type) {
+            case Type::Version: {
+                Response::Version version;
+                std::memcpy(&version, &receive_buffer[sizeof(Header)], sizeof(Response::Version));
+                callback.version(std::move(version));
+                break;
+            }
+            case Type::PortInfo: {
+                Response::PortInfo port_info;
+                std::memcpy(&port_info, &receive_buffer[sizeof(Header)],
+                            sizeof(Response::PortInfo));
+                callback.port_info(std::move(port_info));
+                break;
+            }
+            case Type::PadData: {
+                Response::PadData pad_data;
+                std::memcpy(&pad_data, &receive_buffer[sizeof(Header)], sizeof(Response::PadData));
+                callback.pad_data(std::move(pad_data));
+                break;
+            }
+            }
+        }
+        StartReceive();
+    }
+
+    void HandleSend(const boost::system::error_code& error) {
+        // Send a request for getting port info for the pad
+        Request::PortInfo port_info{1, {pad_index, 0, 0, 0}};
+        const auto port_message = Request::Create(port_info, client_id);
+        std::memcpy(&send_buffer1, &port_message, PORT_INFO_SIZE);
+        socket.send_to(boost::asio::buffer(send_buffer1), send_endpoint);
+
+        // Send a request for getting pad data for the pad
+        Request::PadData pad_data{Request::PadData::Flags::Id, pad_index, EMPTY_MAC_ADDRESS};
+        const auto pad_message = Request::Create(pad_data, client_id);
+        std::memcpy(send_buffer2.data(), &pad_message, PAD_DATA_SIZE);
+        socket.send_to(boost::asio::buffer(send_buffer2), send_endpoint);
+        StartSend(timer.expiry());
+    }
+
+    SocketCallback callback;
+    boost::asio::io_service io_service;
+    boost::asio::basic_waitable_timer<clock> timer;
+    udp::socket socket;
+
+    u32 client_id{};
+    u8 pad_index{};
+
+    static constexpr std::size_t PORT_INFO_SIZE = sizeof(Message<Request::PortInfo>);
+    static constexpr std::size_t PAD_DATA_SIZE = sizeof(Message<Request::PadData>);
+    std::array<u8, PORT_INFO_SIZE> send_buffer1;
+    std::array<u8, PAD_DATA_SIZE> send_buffer2;
+    udp::endpoint send_endpoint;
+
+    std::array<u8, MAX_PACKET_SIZE> receive_buffer;
+    udp::endpoint receive_endpoint;
+};
+
+static void SocketLoop(Socket* socket) {
+    socket->StartReceive();
+    socket->StartSend(Socket::clock::now());
+    socket->Loop();
+}
+
+Client::Client(std::shared_ptr<DeviceStatus> status, const std::string& host, u16 port,
+               u8 pad_index, u32 client_id)
+    : status(status) {
+    StartCommunication(host, port, pad_index, client_id);
+}
+
+Client::~Client() {
+    socket->Stop();
+    thread.join();
+}
+
+void Client::ReloadSocket(const std::string& host, u16 port, u8 pad_index, u32 client_id) {
+    socket->Stop();
+    thread.join();
+    StartCommunication(host, port, pad_index, client_id);
+}
+
+void Client::OnVersion(Response::Version data) {
+    LOG_TRACE(Input, "Version packet received: {}", data.version);
+}
+
+void Client::OnPortInfo(Response::PortInfo data) {
+    LOG_TRACE(Input, "PortInfo packet received: {}", data.model);
+}
+
+void Client::OnPadData(Response::PadData data) {
+    LOG_TRACE(Input, "PadData packet received");
+    if (data.packet_counter <= packet_sequence) {
+        LOG_WARNING(
+            Input,
+            "PadData packet dropped because its stale info. Current count: {} Packet count: {}",
+            packet_sequence, data.packet_counter);
+        return;
+    }
+    packet_sequence = data.packet_counter;
+    // TODO: Check how the Switch handles motions and how the CemuhookUDP motion
+    // directions correspond to the ones of the Switch
+    Common::Vec3f accel = Common::MakeVec<float>(data.accel.x, data.accel.y, data.accel.z);
+    Common::Vec3f gyro = Common::MakeVec<float>(data.gyro.pitch, data.gyro.yaw, data.gyro.roll);
+    {
+        std::lock_guard guard(status->update_mutex);
+
+        status->motion_status = {accel, gyro};
+
+        // TODO: add a setting for "click" touch. Click touch refers to a device that differentiates
+        // between a simple "tap" and a hard press that causes the touch screen to click.
+        const bool is_active = data.touch_1.is_active != 0;
+
+        float x = 0;
+        float y = 0;
+
+        if (is_active && status->touch_calibration) {
+            const u16 min_x = status->touch_calibration->min_x;
+            const u16 max_x = status->touch_calibration->max_x;
+            const u16 min_y = status->touch_calibration->min_y;
+            const u16 max_y = status->touch_calibration->max_y;
+
+            x = (std::clamp(static_cast<u16>(data.touch_1.x), min_x, max_x) - min_x) /
+                static_cast<float>(max_x - min_x);
+            y = (std::clamp(static_cast<u16>(data.touch_1.y), min_y, max_y) - min_y) /
+                static_cast<float>(max_y - min_y);
+        }
+
+        status->touch_status = {x, y, is_active};
+    }
+}
+
+void Client::StartCommunication(const std::string& host, u16 port, u8 pad_index, u32 client_id) {
+    SocketCallback callback{[this](Response::Version version) { OnVersion(version); },
+                            [this](Response::PortInfo info) { OnPortInfo(info); },
+                            [this](Response::PadData data) { OnPadData(data); }};
+    LOG_INFO(Input, "Starting communication with UDP input server on {}:{}", host, port);
+    socket = std::make_unique<Socket>(host, port, pad_index, client_id, callback);
+    thread = std::thread{SocketLoop, this->socket.get()};
+}
+
+void TestCommunication(const std::string& host, u16 port, u8 pad_index, u32 client_id,
+                       std::function<void()> success_callback,
+                       std::function<void()> failure_callback) {
+    std::thread([=] {
+        Common::Event success_event;
+        SocketCallback callback{[](Response::Version version) {}, [](Response::PortInfo info) {},
+                                [&](Response::PadData data) { success_event.Set(); }};
+        Socket socket{host, port, pad_index, client_id, callback};
+        std::thread worker_thread{SocketLoop, &socket};
+        bool result = success_event.WaitFor(std::chrono::seconds(8));
+        socket.Stop();
+        worker_thread.join();
+        if (result) {
+            success_callback();
+        } else {
+            failure_callback();
+        }
+    })
+        .detach();
+}
+
+CalibrationConfigurationJob::CalibrationConfigurationJob(
+    const std::string& host, u16 port, u8 pad_index, u32 client_id,
+    std::function<void(Status)> status_callback,
+    std::function<void(u16, u16, u16, u16)> data_callback) {
+
+    std::thread([=] {
+        constexpr u16 CALIBRATION_THRESHOLD = 100;
+
+        u16 min_x{UINT16_MAX};
+        u16 min_y{UINT16_MAX};
+        u16 max_x{};
+        u16 max_y{};
+
+        Status current_status{Status::Initialized};
+        SocketCallback callback{[](Response::Version version) {}, [](Response::PortInfo info) {},
+                                [&](Response::PadData data) {
+                                    if (current_status == Status::Initialized) {
+                                        // Receiving data means the communication is ready now
+                                        current_status = Status::Ready;
+                                        status_callback(current_status);
+                                    }
+                                    if (!data.touch_1.is_active) {
+                                        return;
+                                    }
+                                    LOG_DEBUG(Input, "Current touch: {} {}", data.touch_1.x,
+                                              data.touch_1.y);
+                                    min_x = std::min(min_x, static_cast<u16>(data.touch_1.x));
+                                    min_y = std::min(min_y, static_cast<u16>(data.touch_1.y));
+                                    if (current_status == Status::Ready) {
+                                        // First touch - min data (min_x/min_y)
+                                        current_status = Status::Stage1Completed;
+                                        status_callback(current_status);
+                                    }
+                                    if (data.touch_1.x - min_x > CALIBRATION_THRESHOLD &&
+                                        data.touch_1.y - min_y > CALIBRATION_THRESHOLD) {
+                                        // Set the current position as max value and finishes
+                                        // configuration
+                                        max_x = data.touch_1.x;
+                                        max_y = data.touch_1.y;
+                                        current_status = Status::Completed;
+                                        data_callback(min_x, min_y, max_x, max_y);
+                                        status_callback(current_status);
+
+                                        complete_event.Set();
+                                    }
+                                }};
+        Socket socket{host, port, pad_index, client_id, callback};
+        std::thread worker_thread{SocketLoop, &socket};
+        complete_event.Wait();
+        socket.Stop();
+        worker_thread.join();
+    })
+        .detach();
+}
+
+CalibrationConfigurationJob::~CalibrationConfigurationJob() {
+    Stop();
+}
+
+void CalibrationConfigurationJob::Stop() {
+    complete_event.Set();
+}
+
+} // namespace InputCommon::CemuhookUDP

src/input_common/udp/client.h

@@ -0,0 +1,96 @@
+// Copyright 2018 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <functional>
+#include <memory>
+#include <mutex>
+#include <optional>
+#include <string>
+#include <thread>
+#include <tuple>
+#include <vector>
+#include "common/common_types.h"
+#include "common/thread.h"
+#include "common/vector_math.h"
+
+namespace InputCommon::CemuhookUDP {
+
+constexpr u16 DEFAULT_PORT = 26760;
+constexpr char DEFAULT_ADDR[] = "127.0.0.1";
+
+class Socket;
+
+namespace Response {
+struct PadData;
+struct PortInfo;
+struct Version;
+} // namespace Response
+
+struct DeviceStatus {
+    std::mutex update_mutex;
+    std::tuple<Common::Vec3<float>, Common::Vec3<float>> motion_status;
+    std::tuple<float, float, bool> touch_status;
+
+    // calibration data for scaling the device's touch area to 3ds
+    struct CalibrationData {
+        u16 min_x{};
+        u16 min_y{};
+        u16 max_x{};
+        u16 max_y{};
+    };
+    std::optional<CalibrationData> touch_calibration;
+};
+
+class Client {
+public:
+    explicit Client(std::shared_ptr<DeviceStatus> status, const std::string& host = DEFAULT_ADDR,
+                    u16 port = DEFAULT_PORT, u8 pad_index = 0, u32 client_id = 24872);
+    ~Client();
+    void ReloadSocket(const std::string& host = "127.0.0.1", u16 port = 26760, u8 pad_index = 0,
+                      u32 client_id = 24872);
+
+private:
+    void OnVersion(Response::Version);
+    void OnPortInfo(Response::PortInfo);
+    void OnPadData(Response::PadData);
+    void StartCommunication(const std::string& host, u16 port, u8 pad_index, u32 client_id);
+
+    std::unique_ptr<Socket> socket;
+    std::shared_ptr<DeviceStatus> status;
+    std::thread thread;
+    u64 packet_sequence = 0;
+};
+
+/// An async job allowing configuration of the touchpad calibration.
+class CalibrationConfigurationJob {
+public:
+    enum class Status {
+        Initialized,
+        Ready,
+        Stage1Completed,
+        Completed,
+    };
+    /**
+     * Constructs and starts the job with the specified parameter.
+     *
+     * @param status_callback Callback for job status updates
+     * @param data_callback Called when calibration data is ready
+     */
+    explicit CalibrationConfigurationJob(const std::string& host, u16 port, u8 pad_index,
+                                         u32 client_id, std::function<void(Status)> status_callback,
+                                         std::function<void(u16, u16, u16, u16)> data_callback);
+    ~CalibrationConfigurationJob();
+    void Stop();
+
+private:
+    Common::Event complete_event;
+};
+
+void TestCommunication(const std::string& host, u16 port, u8 pad_index, u32 client_id,
+                       std::function<void()> success_callback,
+                       std::function<void()> failure_callback);
+
+} // namespace InputCommon::CemuhookUDP

src/input_common/udp/protocol.cpp

@@ -0,0 +1,79 @@
+// Copyright 2018 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <cstddef>
+#include <cstring>
+#include "common/logging/log.h"
+#include "input_common/udp/protocol.h"
+
+namespace InputCommon::CemuhookUDP {
+
+static constexpr std::size_t GetSizeOfResponseType(Type t) {
+    switch (t) {
+    case Type::Version:
+        return sizeof(Response::Version);
+    case Type::PortInfo:
+        return sizeof(Response::PortInfo);
+    case Type::PadData:
+        return sizeof(Response::PadData);
+    }
+    return 0;
+}
+
+namespace Response {
+
+/**
+ * Returns Type if the packet is valid, else none
+ *
+ * Note: Modifies the buffer to zero out the crc (since thats the easiest way to check without
+ * copying the buffer)
+ */
+std::optional<Type> Validate(u8* data, std::size_t size) {
+    if (size < sizeof(Header)) {
+        LOG_DEBUG(Input, "Invalid UDP packet received");
+        return std::nullopt;
+    }
+    Header header{};
+    std::memcpy(&header, data, sizeof(Header));
+    if (header.magic != SERVER_MAGIC) {
+        LOG_ERROR(Input, "UDP Packet has an unexpected magic value");
+        return std::nullopt;
+    }
+    if (header.protocol_version != PROTOCOL_VERSION) {
+        LOG_ERROR(Input, "UDP Packet protocol mismatch");
+        return std::nullopt;
+    }
+    if (header.type < Type::Version || header.type > Type::PadData) {
+        LOG_ERROR(Input, "UDP Packet is an unknown type");
+        return std::nullopt;
+    }
+
+    // Packet size must equal sizeof(Header) + sizeof(Data)
+    // and also verify that the packet info mentions the correct size. Since the spec includes the
+    // type of the packet as part of the data, we need to include it in size calculations here
+    // ie: payload_length == sizeof(T) + sizeof(Type)
+    const std::size_t data_len = GetSizeOfResponseType(header.type);
+    if (header.payload_length != data_len + sizeof(Type) || size < data_len + sizeof(Header)) {
+        LOG_ERROR(
+            Input,
+            "UDP Packet payload length doesn't match. Received: {} PayloadLength: {} Expected: {}",
+            size, header.payload_length, data_len + sizeof(Type));
+        return std::nullopt;
+    }
+
+    const u32 crc32 = header.crc;
+    boost::crc_32_type result;
+    // zero out the crc in the buffer and then run the crc against it
+    std::memset(&data[offsetof(Header, crc)], 0, sizeof(u32_le));
+
+    result.process_bytes(data, data_len + sizeof(Header));
+    if (crc32 != result.checksum()) {
+        LOG_ERROR(Input, "UDP Packet CRC check failed. Offset: {}", offsetof(Header, crc));
+        return std::nullopt;
+    }
+    return header.type;
+}
+} // namespace Response
+
+} // namespace InputCommon::CemuhookUDP

src/input_common/udp/protocol.h

@@ -0,0 +1,256 @@
+// Copyright 2018 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <optional>
+#include <type_traits>
+#include <vector>
+#include <boost/crc.hpp>
+#include "common/bit_field.h"
+#include "common/swap.h"
+
+namespace InputCommon::CemuhookUDP {
+
+constexpr std::size_t MAX_PACKET_SIZE = 100;
+constexpr u16 PROTOCOL_VERSION = 1001;
+constexpr u32 CLIENT_MAGIC = 0x43555344; // DSUC (but flipped for LE)
+constexpr u32 SERVER_MAGIC = 0x53555344; // DSUS (but flipped for LE)
+
+enum class Type : u32 {
+    Version = 0x00100000,
+    PortInfo = 0x00100001,
+    PadData = 0x00100002,
+};
+
+struct Header {
+    u32_le magic{};
+    u16_le protocol_version{};
+    u16_le payload_length{};
+    u32_le crc{};
+    u32_le id{};
+    ///> In the protocol, the type of the packet is not part of the header, but its convenient to
+    ///> include in the header so the callee doesn't have to duplicate the type twice when building
+    ///> the data
+    Type type{};
+};
+static_assert(sizeof(Header) == 20, "UDP Message Header struct has wrong size");
+static_assert(std::is_trivially_copyable_v<Header>, "UDP Message Header is not trivially copyable");
+
+using MacAddress = std::array<u8, 6>;
+constexpr MacAddress EMPTY_MAC_ADDRESS = {0, 0, 0, 0, 0, 0};
+
+#pragma pack(push, 1)
+template <typename T>
+struct Message {
+    Header header{};
+    T data;
+};
+#pragma pack(pop)
+
+template <typename T>
+constexpr Type GetMessageType();
+
+namespace Request {
+
+struct Version {};
+/**
+ * Requests the server to send information about what controllers are plugged into the ports
+ * In citra's case, we only have one controller, so for simplicity's sake, we can just send a
+ * request explicitly for the first controller port and leave it at that. In the future it would be
+ * nice to make this configurable
+ */
+constexpr u32 MAX_PORTS = 4;
+struct PortInfo {
+    u32_le pad_count{}; ///> Number of ports to request data for
+    std::array<u8, MAX_PORTS> port;
+};
+static_assert(std::is_trivially_copyable_v<PortInfo>,
+              "UDP Request PortInfo is not trivially copyable");
+
+/**
+ * Request the latest pad information from the server. If the server hasn't received this message
+ * from the client in a reasonable time frame, the server will stop sending updates. The default
+ * timeout seems to be 5 seconds.
+ */
+struct PadData {
+    enum class Flags : u8 {
+        AllPorts,
+        Id,
+        Mac,
+    };
+    /// Determines which method will be used as a look up for the controller
+    Flags flags{};
+    /// Index of the port of the controller to retrieve data about
+    u8 port_id{};
+    /// Mac address of the controller to retrieve data about
+    MacAddress mac;
+};
+static_assert(sizeof(PadData) == 8, "UDP Request PadData struct has wrong size");
+static_assert(std::is_trivially_copyable_v<PadData>,
+              "UDP Request PadData is not trivially copyable");
+
+/**
+ * Creates a message with the proper header data that can be sent to the server.
+ * @param T data Request body to send
+ * @param client_id ID of the udp client (usually not checked on the server)
+ */
+template <typename T>
+Message<T> Create(const T data, const u32 client_id = 0) {
+    boost::crc_32_type crc;
+    Header header{
+        CLIENT_MAGIC, PROTOCOL_VERSION, sizeof(T) + sizeof(Type), 0, client_id, GetMessageType<T>(),
+    };
+    Message<T> message{header, data};
+    crc.process_bytes(&message, sizeof(Message<T>));
+    message.header.crc = crc.checksum();
+    return message;
+}
+} // namespace Request
+
+namespace Response {
+
+struct Version {
+    u16_le version{};
+};
+static_assert(sizeof(Version) == 2, "UDP Response Version struct has wrong size");
+static_assert(std::is_trivially_copyable_v<Version>,
+              "UDP Response Version is not trivially copyable");
+
+struct PortInfo {
+    u8 id{};
+    u8 state{};
+    u8 model{};
+    u8 connection_type{};
+    MacAddress mac;
+    u8 battery{};
+    u8 is_pad_active{};
+};
+static_assert(sizeof(PortInfo) == 12, "UDP Response PortInfo struct has wrong size");
+static_assert(std::is_trivially_copyable_v<PortInfo>,
+              "UDP Response PortInfo is not trivially copyable");
+
+#pragma pack(push, 1)
+struct PadData {
+    PortInfo info{};
+    u32_le packet_counter{};
+
+    u16_le digital_button{};
+    // The following union isn't trivially copyable but we don't use this input anyway.
+    // union DigitalButton {
+    //     u16_le button;
+    //     BitField<0, 1, u16> button_1;   // Share
+    //     BitField<1, 1, u16> button_2;   // L3
+    //     BitField<2, 1, u16> button_3;   // R3
+    //     BitField<3, 1, u16> button_4;   // Options
+    //     BitField<4, 1, u16> button_5;   // Up
+    //     BitField<5, 1, u16> button_6;   // Right
+    //     BitField<6, 1, u16> button_7;   // Down
+    //     BitField<7, 1, u16> button_8;   // Left
+    //     BitField<8, 1, u16> button_9;   // L2
+    //     BitField<9, 1, u16> button_10;  // R2
+    //     BitField<10, 1, u16> button_11; // L1
+    //     BitField<11, 1, u16> button_12; // R1
+    //     BitField<12, 1, u16> button_13; // Triangle
+    //     BitField<13, 1, u16> button_14; // Circle
+    //     BitField<14, 1, u16> button_15; // Cross
+    //     BitField<15, 1, u16> button_16; // Square
+    // } digital_button;
+
+    u8 home;
+    /// If the device supports a "click" on the touchpad, this will change to 1 when a click happens
+    u8 touch_hard_press{};
+    u8 left_stick_x{};
+    u8 left_stick_y{};
+    u8 right_stick_x{};
+    u8 right_stick_y{};
+
+    struct AnalogButton {
+        u8 button_8{};
+        u8 button_7{};
+        u8 button_6{};
+        u8 button_5{};
+        u8 button_12{};
+        u8 button_11{};
+        u8 button_10{};
+        u8 button_9{};
+        u8 button_16{};
+        u8 button_15{};
+        u8 button_14{};
+        u8 button_13{};
+    } analog_button;
+
+    struct TouchPad {
+        u8 is_active{};
+        u8 id{};
+        u16_le x{};
+        u16_le y{};
+    } touch_1, touch_2;
+
+    u64_le motion_timestamp;
+
+    struct Accelerometer {
+        float x{};
+        float y{};
+        float z{};
+    } accel;
+
+    struct Gyroscope {
+        float pitch{};
+        float yaw{};
+        float roll{};
+    } gyro;
+};
+#pragma pack(pop)
+
+static_assert(sizeof(PadData) == 80, "UDP Response PadData struct has wrong size ");
+static_assert(std::is_trivially_copyable_v<PadData>,
+              "UDP Response PadData is not trivially copyable");
+
+static_assert(sizeof(Message<PadData>) == MAX_PACKET_SIZE,
+              "UDP MAX_PACKET_SIZE is no longer larger than Message<PadData>");
+
+static_assert(sizeof(PadData::AnalogButton) == 12,
+              "UDP Response AnalogButton struct has wrong size ");
+static_assert(sizeof(PadData::TouchPad) == 6, "UDP Response TouchPad struct has wrong size ");
+static_assert(sizeof(PadData::Accelerometer) == 12,
+              "UDP Response Accelerometer struct has wrong size ");
+static_assert(sizeof(PadData::Gyroscope) == 12, "UDP Response Gyroscope struct has wrong size ");
+
+/**
+ * Create a Response Message from the data
+ * @param data array of bytes sent from the server
+ * @return boost::none if it failed to parse or Type if it succeeded. The client can then safely
+ * copy the data into the appropriate struct for that Type
+ */
+std::optional<Type> Validate(u8* data, std::size_t size);
+
+} // namespace Response
+
+template <>
+constexpr Type GetMessageType<Request::Version>() {
+    return Type::Version;
+}
+template <>
+constexpr Type GetMessageType<Request::PortInfo>() {
+    return Type::PortInfo;
+}
+template <>
+constexpr Type GetMessageType<Request::PadData>() {
+    return Type::PadData;
+}
+template <>
+constexpr Type GetMessageType<Response::Version>() {
+    return Type::Version;
+}
+template <>
+constexpr Type GetMessageType<Response::PortInfo>() {
+    return Type::PortInfo;
+}
+template <>
+constexpr Type GetMessageType<Response::PadData>() {
+    return Type::PadData;
+}
+} // namespace InputCommon::CemuhookUDP

src/input_common/udp/udp.cpp

@@ -0,0 +1,96 @@
+// Copyright 2018 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/logging/log.h"
+#include "common/param_package.h"
+#include "core/frontend/input.h"
+#include "core/settings.h"
+#include "input_common/udp/client.h"
+#include "input_common/udp/udp.h"
+
+namespace InputCommon::CemuhookUDP {
+
+class UDPTouchDevice final : public Input::TouchDevice {
+public:
+    explicit UDPTouchDevice(std::shared_ptr<DeviceStatus> status_) : status(std::move(status_)) {}
+    std::tuple<float, float, bool> GetStatus() const {
+        std::lock_guard guard(status->update_mutex);
+        return status->touch_status;
+    }
+
+private:
+    std::shared_ptr<DeviceStatus> status;
+};
+
+class UDPMotionDevice final : public Input::MotionDevice {
+public:
+    explicit UDPMotionDevice(std::shared_ptr<DeviceStatus> status_) : status(std::move(status_)) {}
+    std::tuple<Common::Vec3<float>, Common::Vec3<float>> GetStatus() const {
+        std::lock_guard guard(status->update_mutex);
+        return status->motion_status;
+    }
+
+private:
+    std::shared_ptr<DeviceStatus> status;
+};
+
+class UDPTouchFactory final : public Input::Factory<Input::TouchDevice> {
+public:
+    explicit UDPTouchFactory(std::shared_ptr<DeviceStatus> status_) : status(std::move(status_)) {}
+
+    std::unique_ptr<Input::TouchDevice> Create(const Common::ParamPackage& params) override {
+        {
+            std::lock_guard guard(status->update_mutex);
+            status->touch_calibration.emplace();
+            // These default values work well for DS4 but probably not other touch inputs
+            status->touch_calibration->min_x = params.Get("min_x", 100);
+            status->touch_calibration->min_y = params.Get("min_y", 50);
+            status->touch_calibration->max_x = params.Get("max_x", 1800);
+            status->touch_calibration->max_y = params.Get("max_y", 850);
+        }
+        return std::make_unique<UDPTouchDevice>(status);
+    }
+
+private:
+    std::shared_ptr<DeviceStatus> status;
+};
+
+class UDPMotionFactory final : public Input::Factory<Input::MotionDevice> {
+public:
+    explicit UDPMotionFactory(std::shared_ptr<DeviceStatus> status_) : status(std::move(status_)) {}
+
+    std::unique_ptr<Input::MotionDevice> Create(const Common::ParamPackage& params) override {
+        return std::make_unique<UDPMotionDevice>(status);
+    }
+
+private:
+    std::shared_ptr<DeviceStatus> status;
+};
+
+State::State() {
+    auto status = std::make_shared<DeviceStatus>();
+    client =
+        std::make_unique<Client>(status, Settings::values.udp_input_address,
+                                 Settings::values.udp_input_port, Settings::values.udp_pad_index);
+
+    Input::RegisterFactory<Input::TouchDevice>("cemuhookudp",
+                                               std::make_shared<UDPTouchFactory>(status));
+    Input::RegisterFactory<Input::MotionDevice>("cemuhookudp",
+                                                std::make_shared<UDPMotionFactory>(status));
+}
+
+State::~State() {
+    Input::UnregisterFactory<Input::TouchDevice>("cemuhookudp");
+    Input::UnregisterFactory<Input::MotionDevice>("cemuhookudp");
+}
+
+void State::ReloadUDPClient() {
+    client->ReloadSocket(Settings::values.udp_input_address, Settings::values.udp_input_port,
+                         Settings::values.udp_pad_index);
+}
+
+std::unique_ptr<State> Init() {
+    return std::make_unique<State>();
+}
+} // namespace InputCommon::CemuhookUDP

src/input_common/udp/udp.h

@@ -0,0 +1,27 @@
+// Copyright 2018 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <memory>
+#include <unordered_map>
+#include "input_common/main.h"
+#include "input_common/udp/client.h"
+
+namespace InputCommon::CemuhookUDP {
+
+class UDPTouchDevice;
+class UDPMotionDevice;
+
+class State {
+public:
+    State();
+    ~State();
+    void ReloadUDPClient();
+
+private:
+    std::unique_ptr<Client> client;
+};
+
+std::unique_ptr<State> Init();
+
+} // namespace InputCommon::CemuhookUDP

src/video_core/CMakeLists.txt

@@ -153,14 +153,31 @@ if (ENABLE_VULKAN)
         renderer_vulkan/fixed_pipeline_state.h
         renderer_vulkan/maxwell_to_vk.cpp
         renderer_vulkan/maxwell_to_vk.h
+        renderer_vulkan/renderer_vulkan.h
+        renderer_vulkan/vk_blit_screen.cpp
+        renderer_vulkan/vk_blit_screen.h
         renderer_vulkan/vk_buffer_cache.cpp
         renderer_vulkan/vk_buffer_cache.h
+        renderer_vulkan/vk_compute_pass.cpp
+        renderer_vulkan/vk_compute_pass.h
+        renderer_vulkan/vk_compute_pipeline.cpp
+        renderer_vulkan/vk_compute_pipeline.h
+        renderer_vulkan/vk_descriptor_pool.cpp
+        renderer_vulkan/vk_descriptor_pool.h
         renderer_vulkan/vk_device.cpp
         renderer_vulkan/vk_device.h
+        renderer_vulkan/vk_graphics_pipeline.cpp
+        renderer_vulkan/vk_graphics_pipeline.h
         renderer_vulkan/vk_image.cpp
         renderer_vulkan/vk_image.h
         renderer_vulkan/vk_memory_manager.cpp
         renderer_vulkan/vk_memory_manager.h
+        renderer_vulkan/vk_pipeline_cache.cpp
+        renderer_vulkan/vk_pipeline_cache.h
+        renderer_vulkan/vk_rasterizer.cpp
+        renderer_vulkan/vk_rasterizer.h
+        renderer_vulkan/vk_renderpass_cache.cpp
+        renderer_vulkan/vk_renderpass_cache.h
         renderer_vulkan/vk_resource_manager.cpp
         renderer_vulkan/vk_resource_manager.h
         renderer_vulkan/vk_sampler_cache.cpp
@@ -169,12 +186,19 @@ if (ENABLE_VULKAN)
         renderer_vulkan/vk_scheduler.h
         renderer_vulkan/vk_shader_decompiler.cpp
         renderer_vulkan/vk_shader_decompiler.h
+        renderer_vulkan/vk_shader_util.cpp
+        renderer_vulkan/vk_shader_util.h
         renderer_vulkan/vk_staging_buffer_pool.cpp
         renderer_vulkan/vk_staging_buffer_pool.h
         renderer_vulkan/vk_stream_buffer.cpp
         renderer_vulkan/vk_stream_buffer.h
         renderer_vulkan/vk_swapchain.cpp
-        renderer_vulkan/vk_swapchain.h)
+        renderer_vulkan/vk_swapchain.h
+        renderer_vulkan/vk_texture_cache.cpp
+        renderer_vulkan/vk_texture_cache.h
+        renderer_vulkan/vk_update_descriptor.cpp
+        renderer_vulkan/vk_update_descriptor.h
+    )
 
     target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include)
     target_compile_definitions(video_core PRIVATE HAS_VULKAN)

src/video_core/engines/maxwell_3d.cpp

@@ -91,6 +91,7 @@ void Maxwell3D::InitializeRegisterDefaults() {
     regs.rasterize_enable = 1;
     regs.rt_separate_frag_data = 1;
     regs.framebuffer_srgb = 1;
+    regs.cull.front_face = Maxwell3D::Regs::Cull::FrontFace::ClockWise;
 
     mme_inline[MAXWELL3D_REG_INDEX(draw.vertex_end_gl)] = true;
     mme_inline[MAXWELL3D_REG_INDEX(draw.vertex_begin_gl)] = true;

src/video_core/engines/maxwell_3d.h

@@ -1018,7 +1018,14 @@ public:
                     }
                 } instanced_arrays;
 
-                INSERT_UNION_PADDING_WORDS(0x6);
+                INSERT_UNION_PADDING_WORDS(0x4);
+
+                union {
+                    BitField<0, 1, u32> enable;
+                    BitField<4, 8, u32> unk4;
+                } vp_point_size;
+
+                INSERT_UNION_PADDING_WORDS(1);
 
                 Cull cull;
 
@@ -1271,8 +1278,6 @@ public:
 
     } dirty{};
 
-    std::array<u8, Regs::NUM_REGS> dirty_pointers{};
-
     /// Reads a register value located at the input method address
     u32 GetRegisterValue(u32 method) const;
 
@@ -1367,6 +1372,8 @@ private:
 
     bool execute_on{true};
 
+    std::array<u8, Regs::NUM_REGS> dirty_pointers{};
+
     /// Retrieves information about a specific TIC entry from the TIC buffer.
     Texture::TICEntry GetTICEntry(u32 tic_index) const;
 
@@ -1503,6 +1510,7 @@ ASSERT_REG_POSITION(primitive_restart, 0x591);
 ASSERT_REG_POSITION(index_array, 0x5F2);
 ASSERT_REG_POSITION(polygon_offset_clamp, 0x61F);
 ASSERT_REG_POSITION(instanced_arrays, 0x620);
+ASSERT_REG_POSITION(vp_point_size, 0x644);
 ASSERT_REG_POSITION(cull, 0x646);
 ASSERT_REG_POSITION(pixel_center_integer, 0x649);
 ASSERT_REG_POSITION(viewport_transform_enabled, 0x64B);

src/video_core/engines/shader_bytecode.h

@@ -215,6 +215,18 @@ enum class F2fRoundingOp : u64 {
     Trunc = 11,
 };
 
+enum class AtomicOp : u64 {
+    Add = 0,
+    Min = 1,
+    Max = 2,
+    Inc = 3,
+    Dec = 4,
+    And = 5,
+    Or = 6,
+    Xor = 7,
+    Exch = 8,
+};
+
 enum class UniformType : u64 {
     UnsignedByte = 0,
     SignedByte = 1,
@@ -236,6 +248,13 @@ enum class StoreType : u64 {
     Bits128 = 6,
 };
 
+enum class AtomicType : u64 {
+    U32 = 0,
+    S32 = 1,
+    U64 = 2,
+    S64 = 3,
+};
+
 enum class IMinMaxExchange : u64 {
     None = 0,
     XLo = 1,
@@ -938,6 +957,16 @@ union Instruction {
         BitField<46, 2, u64> cache_mode;
     } stg;
 
+    union {
+        BitField<52, 4, AtomicOp> operation;
+        BitField<28, 2, AtomicType> type;
+        BitField<30, 22, s64> offset;
+
+        s32 GetImmediateOffset() const {
+            return static_cast<s32>(offset << 2);
+        }
+    } atoms;
+
     union {
         BitField<32, 1, PhysicalAttributeDirection> direction;
         BitField<47, 3, AttributeSize> size;
@@ -1659,9 +1688,10 @@ public:
         ST_A,
         ST_L,
         ST_S,
-        ST,   // Store in generic memory
-        STG,  // Store in global memory
-        AL2P, // Transforms attribute memory into physical memory
+        ST,    // Store in generic memory
+        STG,   // Store in global memory
+        ATOMS, // Atomic operation on shared memory
+        AL2P,  // Transforms attribute memory into physical memory
         TEX,
         TEX_B,  // Texture Load Bindless
         TXQ,    // Texture Query
@@ -1964,6 +1994,7 @@ private:
             INST("1110111101010---", Id::ST_L, Type::Memory, "ST_L"),
             INST("101-------------", Id::ST, Type::Memory, "ST"),
             INST("1110111011011---", Id::STG, Type::Memory, "STG"),
+            INST("11101100--------", Id::ATOMS, Type::Memory, "ATOMS"),
             INST("1110111110100---", Id::AL2P, Type::Memory, "AL2P"),
             INST("110000----111---", Id::TEX, Type::Texture, "TEX"),
             INST("1101111010111---", Id::TEX_B, Type::Texture, "TEX_B"),

src/video_core/gpu.cpp

@@ -66,19 +66,20 @@ const DmaPusher& GPU::DmaPusher() const {
     return *dma_pusher;
 }
 
-void GPU::WaitFence(u32 syncpoint_id, u32 value) const {
+void GPU::WaitFence(u32 syncpoint_id, u32 value) {
     // Synced GPU, is always in sync
     if (!is_async) {
         return;
     }
     MICROPROFILE_SCOPE(GPU_wait);
-    while (syncpoints[syncpoint_id].load(std::memory_order_relaxed) < value) {
-    }
+    std::unique_lock lock{sync_mutex};
+    sync_cv.wait(lock, [=]() { return syncpoints[syncpoint_id].load() >= value; });
 }
 
 void GPU::IncrementSyncPoint(const u32 syncpoint_id) {
     syncpoints[syncpoint_id]++;
     std::lock_guard lock{sync_mutex};
+    sync_cv.notify_all();
     if (!syncpt_interrupts[syncpoint_id].empty()) {
         u32 value = syncpoints[syncpoint_id].load();
         auto it = syncpt_interrupts[syncpoint_id].begin();

src/video_core/gpu.h

@@ -6,6 +6,7 @@
 
 #include <array>
 #include <atomic>
+#include <condition_variable>
 #include <list>
 #include <memory>
 #include <mutex>
@@ -181,7 +182,7 @@ public:
     virtual void WaitIdle() const = 0;
 
     /// Allows the CPU/NvFlinger to wait on the GPU before presenting a frame.
-    void WaitFence(u32 syncpoint_id, u32 value) const;
+    void WaitFence(u32 syncpoint_id, u32 value);
 
     void IncrementSyncPoint(u32 syncpoint_id);
 
@@ -312,6 +313,8 @@ private:
 
     std::mutex sync_mutex;
 
+    std::condition_variable sync_cv;
+
     const bool is_async;
 };
 

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -1272,6 +1272,7 @@ void RasterizerOpenGL::SyncPointState() {
     const auto& regs = system.GPU().Maxwell3D().regs;
     // Limit the point size to 1 since nouveau sometimes sets a point size of 0 (and that's invalid
     // in OpenGL).
+    state.point.program_control = regs.vp_point_size.enable != 0;
     state.point.size = std::max(1.0f, regs.point_size);
 }
 

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -34,9 +34,6 @@ using VideoCommon::Shader::ShaderIR;
 
 namespace {
 
-// One UBO is always reserved for emulation values on staged shaders
-constexpr u32 STAGE_RESERVED_UBOS = 1;
-
 constexpr u32 STAGE_MAIN_OFFSET = 10;
 constexpr u32 KERNEL_MAIN_OFFSET = 0;
 
@@ -243,7 +240,6 @@ CachedProgram BuildShader(const Device& device, u64 unique_identifier, ShaderTyp
     if (!code_b.empty()) {
         ir_b.emplace(code_b, main_offset, COMPILER_SETTINGS, locker);
     }
-    const auto entries = GLShader::GetEntries(ir);
 
     std::string source = fmt::format(R"(// {}
 #version 430 core
@@ -264,6 +260,10 @@ CachedProgram BuildShader(const Device& device, u64 unique_identifier, ShaderTyp
                   "#extension GL_NV_shader_thread_group : require\n"
                   "#extension GL_NV_shader_thread_shuffle : require\n";
     }
+    // This pragma stops Nvidia's driver from over optimizing math (probably using fp16 operations)
+    // on places where we don't want to.
+    // Thanks to Ryujinx for finding this workaround.
+    source += "#pragma optionNV(fastmath off)\n";
 
     if (shader_type == ShaderType::Geometry) {
         const auto [glsl_topology, max_vertices] = GetPrimitiveDescription(variant.primitive_mode);
@@ -314,9 +314,10 @@ std::unordered_set<GLenum> GetSupportedFormats() {
 
 CachedShader::CachedShader(const ShaderParameters& params, ShaderType shader_type,
                            GLShader::ShaderEntries entries, ProgramCode code, ProgramCode code_b)
-    : RasterizerCacheObject{params.host_ptr}, system{params.system}, disk_cache{params.disk_cache},
-      device{params.device}, cpu_addr{params.cpu_addr}, unique_identifier{params.unique_identifier},
-      shader_type{shader_type}, entries{entries}, code{std::move(code)}, code_b{std::move(code_b)} {
+    : RasterizerCacheObject{params.host_ptr}, system{params.system},
+      disk_cache{params.disk_cache}, device{params.device}, cpu_addr{params.cpu_addr},
+      unique_identifier{params.unique_identifier}, shader_type{shader_type},
+      entries{std::move(entries)}, code{std::move(code)}, code_b{std::move(code_b)} {
     if (!params.precompiled_variants) {
         return;
     }

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -751,6 +751,9 @@ private:
 
     Expression Visit(const Node& node) {
         if (const auto operation = std::get_if<OperationNode>(&*node)) {
+            if (const auto amend_index = operation->GetAmendIndex()) {
+                Visit(ir.GetAmendNode(*amend_index)).CheckVoid();
+            }
             const auto operation_index = static_cast<std::size_t>(operation->GetCode());
             if (operation_index >= operation_decompilers.size()) {
                 UNREACHABLE_MSG("Out of bounds operation: {}", operation_index);
@@ -872,6 +875,9 @@ private:
         }
 
         if (const auto conditional = std::get_if<ConditionalNode>(&*node)) {
+            if (const auto amend_index = conditional->GetAmendIndex()) {
+                Visit(ir.GetAmendNode(*amend_index)).CheckVoid();
+            }
             // It's invalid to call conditional on nested nodes, use an operation instead
             code.AddLine("if ({}) {{", Visit(conditional->GetCondition()).AsBool());
             ++code.scope;
@@ -1850,6 +1856,16 @@ private:
                 Type::Uint};
     }
 
+    template <const std::string_view& opname, Type type>
+    Expression Atomic(Operation operation) {
+        ASSERT(stage == ShaderType::Compute);
+        auto& smem = std::get<SmemNode>(*operation[0]);
+
+        return {fmt::format("atomic{}(smem[{} >> 2], {})", opname, Visit(smem.GetAddress()).AsInt(),
+                            Visit(operation[1]).As(type)),
+                type};
+    }
+
     Expression Branch(Operation operation) {
         const auto target = std::get_if<ImmediateNode>(&*operation[0]);
         UNIMPLEMENTED_IF(!target);
@@ -2188,6 +2204,8 @@ private:
         &GLSLDecompiler::AtomicImage<Func::Xor>,
         &GLSLDecompiler::AtomicImage<Func::Exchange>,
 
+        &GLSLDecompiler::Atomic<Func::Add, Type::Uint>,
+
         &GLSLDecompiler::Branch,
         &GLSLDecompiler::BranchIndirect,
         &GLSLDecompiler::PushFlowStack,
@@ -2307,7 +2325,7 @@ public:
     explicit ExprDecompiler(GLSLDecompiler& decomp) : decomp{decomp} {}
 
     void operator()(const ExprAnd& expr) {
-        inner += "( ";
+        inner += '(';
         std::visit(*this, *expr.operand1);
         inner += " && ";
         std::visit(*this, *expr.operand2);
@@ -2315,7 +2333,7 @@ public:
     }
 
     void operator()(const ExprOr& expr) {
-        inner += "( ";
+        inner += '(';
         std::visit(*this, *expr.operand1);
         inner += " || ";
         std::visit(*this, *expr.operand2);
@@ -2333,28 +2351,7 @@ public:
     }
 
     void operator()(const ExprCondCode& expr) {
-        const Node cc = decomp.ir.GetConditionCode(expr.cc);
-        std::string target;
-
-        if (const auto pred = std::get_if<PredicateNode>(&*cc)) {
-            const auto index = pred->GetIndex();
-            switch (index) {
-            case Tegra::Shader::Pred::NeverExecute:
-                target = "false";
-                break;
-            case Tegra::Shader::Pred::UnusedIndex:
-                target = "true";
-                break;
-            default:
-                target = decomp.GetPredicate(index);
-                break;
-            }
-        } else if (const auto flag = std::get_if<InternalFlagNode>(&*cc)) {
-            target = decomp.GetInternalFlag(flag->GetFlag());
-        } else {
-            UNREACHABLE();
-        }
-        inner += target;
+        inner += decomp.Visit(decomp.ir.GetConditionCode(expr.cc)).AsBool();
     }
 
     void operator()(const ExprVar& expr) {
@@ -2366,8 +2363,7 @@ public:
     }
 
     void operator()(VideoCommon::Shader::ExprGprEqual& expr) {
-        inner +=
-            "( ftou(" + decomp.GetRegister(expr.gpr) + ") == " + std::to_string(expr.value) + ')';
+        inner += fmt::format("(ftou({}) == {})", decomp.GetRegister(expr.gpr), expr.value);
     }
 
     const std::string& GetResult() const {
@@ -2375,8 +2371,8 @@ public:
     }
 
 private:
-    std::string inner;
     GLSLDecompiler& decomp;
+    std::string inner;
 };
 
 class ASTDecompiler {

src/video_core/renderer_opengl/gl_state.cpp

@@ -127,6 +127,7 @@ void OpenGLState::ApplyClipDistances() {
 }
 
 void OpenGLState::ApplyPointSize() {
+    Enable(GL_PROGRAM_POINT_SIZE, cur_state.point.program_control, point.program_control);
     if (UpdateValue(cur_state.point.size, point.size)) {
         glPointSize(point.size);
     }

src/video_core/renderer_opengl/gl_state.h

@@ -131,7 +131,8 @@ public:
     std::array<Viewport, Tegra::Engines::Maxwell3D::Regs::NumViewports> viewports;
 
     struct {
-        float size = 1.0f; // GL_POINT_SIZE
+        bool program_control = false; // GL_PROGRAM_POINT_SIZE
+        GLfloat size = 1.0f;          // GL_POINT_SIZE
     } point;
 
     struct {

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -44,7 +44,7 @@ struct FormatTuple {
 
 constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex_format_tuples = {{
     {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, false},                        // ABGR8U
-    {GL_RGBA8, GL_RGBA, GL_BYTE, false},                                            // ABGR8S
+    {GL_RGBA8_SNORM, GL_RGBA, GL_BYTE, false},                                      // ABGR8S
     {GL_RGBA8UI, GL_RGBA_INTEGER, GL_UNSIGNED_BYTE, false},                         // ABGR8UI
     {GL_RGB565, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV, false},                        // B5G6R5U
     {GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, false},                  // A2B10G10R10U
@@ -83,9 +83,9 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex_format
     {GL_RGB32F, GL_RGB, GL_FLOAT, false},                                           // RGB32F
     {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, false},                 // RGBA8_SRGB
     {GL_RG8, GL_RG, GL_UNSIGNED_BYTE, false},                                       // RG8U
-    {GL_RG8, GL_RG, GL_BYTE, false},                                                // RG8S
+    {GL_RG8_SNORM, GL_RG, GL_BYTE, false},                                          // RG8S
     {GL_RG32UI, GL_RG_INTEGER, GL_UNSIGNED_INT, false},                             // RG32UI
-    {GL_RGB16F, GL_RGBA16, GL_HALF_FLOAT, false},                                   // RGBX16F
+    {GL_RGB16F, GL_RGBA, GL_HALF_FLOAT, false},                                     // RGBX16F
     {GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT, false},                             // R32UI
     {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, false},                                   // ASTC_2D_8X8
     {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, false},                                   // ASTC_2D_8X5
@@ -253,14 +253,12 @@ void CachedSurface::DownloadTexture(std::vector<u8>& staging_buffer) {
         glPixelStorei(GL_PACK_ALIGNMENT, std::min(8U, params.GetRowAlignment(level)));
         glPixelStorei(GL_PACK_ROW_LENGTH, static_cast<GLint>(params.GetMipWidth(level)));
         const std::size_t mip_offset = params.GetHostMipmapLevelOffset(level);
+        u8* const mip_data = staging_buffer.data() + mip_offset;
+        const GLsizei size = static_cast<GLsizei>(params.GetHostMipmapSize(level));
         if (is_compressed) {
-            glGetCompressedTextureImage(texture.handle, level,
-                                        static_cast<GLsizei>(params.GetHostMipmapSize(level)),
-                                        staging_buffer.data() + mip_offset);
+            glGetCompressedTextureImage(texture.handle, level, size, mip_data);
         } else {
-            glGetTextureImage(texture.handle, level, format, type,
-                              static_cast<GLsizei>(params.GetHostMipmapSize(level)),
-                              staging_buffer.data() + mip_offset);
+            glGetTextureImage(texture.handle, level, format, type, size, mip_data);
         }
     }
 }

src/video_core/renderer_opengl/utils.cpp

@@ -6,16 +6,20 @@
 #include <vector>
 
 #include <fmt/format.h>
-
 #include <glad/glad.h>
 
-#include "common/assert.h"
 #include "common/common_types.h"
-#include "common/scope_exit.h"
 #include "video_core/renderer_opengl/utils.h"
 
 namespace OpenGL {
 
+struct VertexArrayPushBuffer::Entry {
+    GLuint binding_index{};
+    const GLuint* buffer{};
+    GLintptr offset{};
+    GLsizei stride{};
+};
+
 VertexArrayPushBuffer::VertexArrayPushBuffer() = default;
 
 VertexArrayPushBuffer::~VertexArrayPushBuffer() = default;
@@ -47,6 +51,13 @@ void VertexArrayPushBuffer::Bind() {
     }
 }
 
+struct BindBuffersRangePushBuffer::Entry {
+    GLuint binding;
+    const GLuint* buffer;
+    GLintptr offset;
+    GLsizeiptr size;
+};
+
 BindBuffersRangePushBuffer::BindBuffersRangePushBuffer(GLenum target) : target{target} {}
 
 BindBuffersRangePushBuffer::~BindBuffersRangePushBuffer() = default;

src/video_core/renderer_opengl/utils.h

@@ -26,12 +26,7 @@ public:
     void Bind();
 
 private:
-    struct Entry {
-        GLuint binding_index{};
-        const GLuint* buffer{};
-        GLintptr offset{};
-        GLsizei stride{};
-    };
+    struct Entry;
 
     GLuint vao{};
     const GLuint* index_buffer{};
@@ -50,12 +45,7 @@ public:
     void Bind();
 
 private:
-    struct Entry {
-        GLuint binding;
-        const GLuint* buffer;
-        GLintptr offset;
-        GLsizeiptr size;
-    };
+    struct Entry;
 
     GLenum target;
     std::vector<Entry> entries;

src/video_core/renderer_vulkan/fixed_pipeline_state.cpp

@@ -109,6 +109,9 @@ constexpr FixedPipelineState::Rasterizer GetRasterizerState(const Maxwell& regs)
     const auto topology = static_cast<std::size_t>(regs.draw.topology.Value());
     const bool depth_bias_enabled = enabled_lut[PolygonOffsetEnableLUT[topology]];
 
+    const auto& clip = regs.view_volume_clip_control;
+    const bool depth_clamp_enabled = clip.depth_clamp_near == 1 || clip.depth_clamp_far == 1;
+
     Maxwell::Cull::FrontFace front_face = regs.cull.front_face;
     if (regs.screen_y_control.triangle_rast_flip != 0 &&
         regs.viewport_transform[0].scale_y > 0.0f) {
@@ -119,8 +122,9 @@ constexpr FixedPipelineState::Rasterizer GetRasterizerState(const Maxwell& regs)
     }
 
     const bool gl_ndc = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne;
-    return FixedPipelineState::Rasterizer(regs.cull.enabled, depth_bias_enabled, gl_ndc,
-                                          regs.cull.cull_face, front_face);
+    return FixedPipelineState::Rasterizer(regs.cull.enabled, depth_bias_enabled,
+                                          depth_clamp_enabled, gl_ndc, regs.cull.cull_face,
+                                          front_face);
 }
 
 } // Anonymous namespace
@@ -222,15 +226,17 @@ bool FixedPipelineState::Tessellation::operator==(const Tessellation& rhs) const
 std::size_t FixedPipelineState::Rasterizer::Hash() const noexcept {
     return static_cast<std::size_t>(cull_enable) ^
            (static_cast<std::size_t>(depth_bias_enable) << 1) ^
-           (static_cast<std::size_t>(ndc_minus_one_to_one) << 2) ^
+           (static_cast<std::size_t>(depth_clamp_enable) << 2) ^
+           (static_cast<std::size_t>(ndc_minus_one_to_one) << 3) ^
            (static_cast<std::size_t>(cull_face) << 24) ^
            (static_cast<std::size_t>(front_face) << 48);
 }
 
 bool FixedPipelineState::Rasterizer::operator==(const Rasterizer& rhs) const noexcept {
-    return std::tie(cull_enable, depth_bias_enable, ndc_minus_one_to_one, cull_face, front_face) ==
-           std::tie(rhs.cull_enable, rhs.depth_bias_enable, rhs.ndc_minus_one_to_one, rhs.cull_face,
-                    rhs.front_face);
+    return std::tie(cull_enable, depth_bias_enable, depth_clamp_enable, ndc_minus_one_to_one,
+                    cull_face, front_face) ==
+           std::tie(rhs.cull_enable, rhs.depth_bias_enable, rhs.depth_clamp_enable,
+                    rhs.ndc_minus_one_to_one, rhs.cull_face, rhs.front_face);
 }
 
 std::size_t FixedPipelineState::DepthStencil::Hash() const noexcept {

src/video_core/renderer_vulkan/fixed_pipeline_state.h

@@ -170,15 +170,17 @@ struct FixedPipelineState {
     };
 
     struct Rasterizer {
-        constexpr Rasterizer(bool cull_enable, bool depth_bias_enable, bool ndc_minus_one_to_one,
-                             Maxwell::Cull::CullFace cull_face, Maxwell::Cull::FrontFace front_face)
+        constexpr Rasterizer(bool cull_enable, bool depth_bias_enable, bool depth_clamp_enable,
+                             bool ndc_minus_one_to_one, Maxwell::Cull::CullFace cull_face,
+                             Maxwell::Cull::FrontFace front_face)
             : cull_enable{cull_enable}, depth_bias_enable{depth_bias_enable},
-              ndc_minus_one_to_one{ndc_minus_one_to_one}, cull_face{cull_face}, front_face{
-                                                                                    front_face} {}
+              depth_clamp_enable{depth_clamp_enable}, ndc_minus_one_to_one{ndc_minus_one_to_one},
+              cull_face{cull_face}, front_face{front_face} {}
         Rasterizer() = default;
 
         bool cull_enable;
         bool depth_bias_enable;
+        bool depth_clamp_enable;
         bool ndc_minus_one_to_one;
         Maxwell::Cull::CullFace cull_face;
         Maxwell::Cull::FrontFace front_face;

src/video_core/renderer_vulkan/maxwell_to_vk.cpp

@@ -44,7 +44,7 @@ vk::SamplerMipmapMode MipmapMode(Tegra::Texture::TextureMipmapFilter mipmap_filt
     return {};
 }
 
-vk::SamplerAddressMode WrapMode(Tegra::Texture::WrapMode wrap_mode,
+vk::SamplerAddressMode WrapMode(const VKDevice& device, Tegra::Texture::WrapMode wrap_mode,
                                 Tegra::Texture::TextureFilter filter) {
     switch (wrap_mode) {
     case Tegra::Texture::WrapMode::Wrap:
@@ -56,7 +56,12 @@ vk::SamplerAddressMode WrapMode(Tegra::Texture::WrapMode wrap_mode,
     case Tegra::Texture::WrapMode::Border:
         return vk::SamplerAddressMode::eClampToBorder;
     case Tegra::Texture::WrapMode::Clamp:
-        // TODO(Rodrigo): Emulate GL_CLAMP properly
+        if (device.GetDriverID() == vk::DriverIdKHR::eNvidiaProprietary) {
+            // Nvidia's Vulkan driver defaults to GL_CLAMP on invalid enumerations, we can hack this
+            // by sending an invalid enumeration.
+            return static_cast<vk::SamplerAddressMode>(0xcafe);
+        }
+        // TODO(Rodrigo): Emulate GL_CLAMP properly on other vendors
         switch (filter) {
         case Tegra::Texture::TextureFilter::Nearest:
             return vk::SamplerAddressMode::eClampToEdge;

src/video_core/renderer_vulkan/maxwell_to_vk.h

@@ -22,7 +22,7 @@ vk::Filter Filter(Tegra::Texture::TextureFilter filter);
 
 vk::SamplerMipmapMode MipmapMode(Tegra::Texture::TextureMipmapFilter mipmap_filter);
 
-vk::SamplerAddressMode WrapMode(Tegra::Texture::WrapMode wrap_mode,
+vk::SamplerAddressMode WrapMode(const VKDevice& device, Tegra::Texture::WrapMode wrap_mode,
                                 Tegra::Texture::TextureFilter filter);
 
 vk::CompareOp DepthCompareFunction(Tegra::Texture::DepthCompareFunc depth_compare_func);

src/video_core/renderer_vulkan/renderer_vulkan.h

@@ -0,0 +1,72 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <optional>
+#include <vector>
+#include "video_core/renderer_base.h"
+#include "video_core/renderer_vulkan/declarations.h"
+
+namespace Core {
+class System;
+}
+
+namespace Vulkan {
+
+class VKBlitScreen;
+class VKDevice;
+class VKFence;
+class VKMemoryManager;
+class VKResourceManager;
+class VKSwapchain;
+class VKScheduler;
+class VKImage;
+
+struct VKScreenInfo {
+    VKImage* image{};
+    u32 width{};
+    u32 height{};
+    bool is_srgb{};
+};
+
+class RendererVulkan final : public VideoCore::RendererBase {
+public:
+    explicit RendererVulkan(Core::Frontend::EmuWindow& window, Core::System& system);
+    ~RendererVulkan() override;
+
+    /// Swap buffers (render frame)
+    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
+
+    /// Initialize the renderer
+    bool Init() override;
+
+    /// Shutdown the renderer
+    void ShutDown() override;
+
+private:
+    std::optional<vk::DebugUtilsMessengerEXT> CreateDebugCallback(
+        const vk::DispatchLoaderDynamic& dldi);
+
+    bool PickDevices(const vk::DispatchLoaderDynamic& dldi);
+
+    void Report() const;
+
+    Core::System& system;
+
+    vk::Instance instance;
+    vk::SurfaceKHR surface;
+
+    VKScreenInfo screen_info;
+
+    UniqueDebugUtilsMessengerEXT debug_callback;
+    std::unique_ptr<VKDevice> device;
+    std::unique_ptr<VKSwapchain> swapchain;
+    std::unique_ptr<VKMemoryManager> memory_manager;
+    std::unique_ptr<VKResourceManager> resource_manager;
+    std::unique_ptr<VKScheduler> scheduler;
+    std::unique_ptr<VKBlitScreen> blit_screen;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_blit_screen.cpp

@@ -0,0 +1,627 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <array>
+#include <cstring>
+#include <memory>
+#include <tuple>
+#include <vector>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "common/math_util.h"
+
+#include "core/core.h"
+#include "core/frontend/emu_window.h"
+#include "core/memory.h"
+
+#include "video_core/gpu.h"
+#include "video_core/morton.h"
+#include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/renderer_vulkan.h"
+#include "video_core/renderer_vulkan/vk_blit_screen.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_image.h"
+#include "video_core/renderer_vulkan/vk_memory_manager.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_shader_util.h"
+#include "video_core/renderer_vulkan/vk_swapchain.h"
+#include "video_core/surface.h"
+
+namespace Vulkan {
+
+namespace {
+
+// Generated from the "shaders/" directory, read the instructions there.
+constexpr u8 blit_vertex_code[] = {
+    0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x27, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
+    0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x0f, 0x00, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e,
+    0x00, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
+    0x25, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x0b, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x0b, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00,
+    0x0b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
+    0x48, 0x00, 0x05, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
+    0x04, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x48, 0x00, 0x04, 0x00, 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05, 0x00, 0x00, 0x00,
+    0x48, 0x00, 0x05, 0x00, 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x07, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x11, 0x00, 0x00, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x19, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x24, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x25, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00,
+    0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x16, 0x00, 0x03, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x20, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x04, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x08, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x08, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x06, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
+    0x0c, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
+    0x0c, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
+    0x0e, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00,
+    0x0e, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x18, 0x00, 0x04, 0x00,
+    0x10, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00,
+    0x11, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x12, 0x00, 0x00, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x12, 0x00, 0x00, 0x00,
+    0x13, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x14, 0x00, 0x00, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00,
+    0x19, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x1b, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x1c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x80, 0x3f, 0x20, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00,
+    0x03, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0x03, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0x24, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00,
+    0x25, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00,
+    0x05, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x14, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
+    0x13, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
+    0x16, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00,
+    0x1a, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x51, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x1d, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x51, 0x00, 0x05, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x50, 0x00, 0x07, 0x00, 0x07, 0x00, 0x00, 0x00, 0x1f, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
+    0x1e, 0x00, 0x00, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x91, 0x00, 0x05, 0x00,
+    0x07, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x1f, 0x00, 0x00, 0x00,
+    0x41, 0x00, 0x05, 0x00, 0x21, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00,
+    0x0f, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x22, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
+    0x3d, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x26, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
+    0x3e, 0x00, 0x03, 0x00, 0x24, 0x00, 0x00, 0x00, 0x26, 0x00, 0x00, 0x00, 0xfd, 0x00, 0x01, 0x00,
+    0x38, 0x00, 0x01, 0x00};
+
+constexpr u8 blit_fragment_code[] = {
+    0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x14, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
+    0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x0f, 0x00, 0x07, 0x00, 0x04, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e,
+    0x00, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x10, 0x00, 0x03, 0x00,
+    0x04, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
+    0x1e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0d, 0x00, 0x00, 0x00,
+    0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0d, 0x00, 0x00, 0x00,
+    0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x11, 0x00, 0x00, 0x00,
+    0x1e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x21, 0x00, 0x03, 0x00, 0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x16, 0x00, 0x03, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x08, 0x00, 0x00, 0x00,
+    0x03, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x08, 0x00, 0x00, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x19, 0x00, 0x09, 0x00, 0x0a, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1b, 0x00, 0x03, 0x00,
+    0x0b, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0c, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x0c, 0x00, 0x00, 0x00,
+    0x0d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x0f, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x10, 0x00, 0x00, 0x00,
+    0x11, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00,
+    0x05, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x00, 0x00,
+    0x0d, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x12, 0x00, 0x00, 0x00,
+    0x11, 0x00, 0x00, 0x00, 0x57, 0x00, 0x05, 0x00, 0x07, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00,
+    0x0e, 0x00, 0x00, 0x00, 0x12, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x09, 0x00, 0x00, 0x00,
+    0x13, 0x00, 0x00, 0x00, 0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00};
+
+struct ScreenRectVertex {
+    ScreenRectVertex() = default;
+    explicit ScreenRectVertex(f32 x, f32 y, f32 u, f32 v) : position{{x, y}}, tex_coord{{u, v}} {}
+
+    std::array<f32, 2> position;
+    std::array<f32, 2> tex_coord;
+
+    static vk::VertexInputBindingDescription GetDescription() {
+        return vk::VertexInputBindingDescription(0, sizeof(ScreenRectVertex),
+                                                 vk::VertexInputRate::eVertex);
+    }
+
+    static std::array<vk::VertexInputAttributeDescription, 2> GetAttributes() {
+        return {vk::VertexInputAttributeDescription(0, 0, vk::Format::eR32G32Sfloat,
+                                                    offsetof(ScreenRectVertex, position)),
+                vk::VertexInputAttributeDescription(1, 0, vk::Format::eR32G32Sfloat,
+                                                    offsetof(ScreenRectVertex, tex_coord))};
+    }
+};
+
+constexpr std::array<f32, 4 * 4> MakeOrthographicMatrix(f32 width, f32 height) {
+    // clang-format off
+    return { 2.f / width, 0.f,          0.f, 0.f,
+             0.f,         2.f / height, 0.f, 0.f,
+             0.f,         0.f,          1.f, 0.f,
+            -1.f,        -1.f,          0.f, 1.f};
+    // clang-format on
+}
+
+std::size_t GetBytesPerPixel(const Tegra::FramebufferConfig& framebuffer) {
+    using namespace VideoCore::Surface;
+    return GetBytesPerPixel(PixelFormatFromGPUPixelFormat(framebuffer.pixel_format));
+}
+
+std::size_t GetSizeInBytes(const Tegra::FramebufferConfig& framebuffer) {
+    return static_cast<std::size_t>(framebuffer.stride) *
+           static_cast<std::size_t>(framebuffer.height) * GetBytesPerPixel(framebuffer);
+}
+
+vk::Format GetFormat(const Tegra::FramebufferConfig& framebuffer) {
+    switch (framebuffer.pixel_format) {
+    case Tegra::FramebufferConfig::PixelFormat::ABGR8:
+        return vk::Format::eA8B8G8R8UnormPack32;
+    case Tegra::FramebufferConfig::PixelFormat::RGB565:
+        return vk::Format::eR5G6B5UnormPack16;
+    default:
+        UNIMPLEMENTED_MSG("Unknown framebuffer pixel format: {}",
+                          static_cast<u32>(framebuffer.pixel_format));
+        return vk::Format::eA8B8G8R8UnormPack32;
+    }
+}
+
+} // Anonymous namespace
+
+struct VKBlitScreen::BufferData {
+    struct {
+        std::array<f32, 4 * 4> modelview_matrix;
+    } uniform;
+
+    std::array<ScreenRectVertex, 4> vertices;
+
+    // Unaligned image data goes here
+};
+
+VKBlitScreen::VKBlitScreen(Core::System& system, Core::Frontend::EmuWindow& render_window,
+                           VideoCore::RasterizerInterface& rasterizer, const VKDevice& device,
+                           VKResourceManager& resource_manager, VKMemoryManager& memory_manager,
+                           VKSwapchain& swapchain, VKScheduler& scheduler,
+                           const VKScreenInfo& screen_info)
+    : system{system}, render_window{render_window}, rasterizer{rasterizer}, device{device},
+      resource_manager{resource_manager}, memory_manager{memory_manager}, swapchain{swapchain},
+      scheduler{scheduler}, image_count{swapchain.GetImageCount()}, screen_info{screen_info} {
+    watches.resize(image_count);
+    std::generate(watches.begin(), watches.end(),
+                  []() { return std::make_unique<VKFenceWatch>(); });
+
+    CreateStaticResources();
+    CreateDynamicResources();
+}
+
+VKBlitScreen::~VKBlitScreen() = default;
+
+void VKBlitScreen::Recreate() {
+    CreateDynamicResources();
+}
+
+std::tuple<VKFence&, vk::Semaphore> VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
+                                                       bool use_accelerated) {
+    RefreshResources(framebuffer);
+
+    // Finish any pending renderpass
+    scheduler.RequestOutsideRenderPassOperationContext();
+
+    const std::size_t image_index = swapchain.GetImageIndex();
+    watches[image_index]->Watch(scheduler.GetFence());
+
+    VKImage* blit_image = use_accelerated ? screen_info.image : raw_images[image_index].get();
+
+    UpdateDescriptorSet(image_index, blit_image->GetPresentView());
+
+    BufferData data;
+    SetUniformData(data, framebuffer);
+    SetVertexData(data, framebuffer);
+
+    auto map = buffer_commit->Map();
+    std::memcpy(map.GetAddress(), &data, sizeof(data));
+
+    if (!use_accelerated) {
+        const u64 image_offset = GetRawImageOffset(framebuffer, image_index);
+
+        const auto pixel_format =
+            VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format);
+        const VAddr framebuffer_addr = framebuffer.address + framebuffer.offset;
+        const auto host_ptr = system.Memory().GetPointer(framebuffer_addr);
+        rasterizer.FlushRegion(ToCacheAddr(host_ptr), GetSizeInBytes(framebuffer));
+
+        // TODO(Rodrigo): Read this from HLE
+        constexpr u32 block_height_log2 = 4;
+        VideoCore::MortonSwizzle(VideoCore::MortonSwizzleMode::MortonToLinear, pixel_format,
+                                 framebuffer.stride, block_height_log2, framebuffer.height, 0, 1, 1,
+                                 map.GetAddress() + image_offset, host_ptr);
+
+        blit_image->Transition(0, 1, 0, 1, vk::PipelineStageFlagBits::eTransfer,
+                               vk::AccessFlagBits::eTransferWrite,
+                               vk::ImageLayout::eTransferDstOptimal);
+
+        const vk::BufferImageCopy copy(image_offset, 0, 0,
+                                       {vk::ImageAspectFlagBits::eColor, 0, 0, 1}, {0, 0, 0},
+                                       {framebuffer.width, framebuffer.height, 1});
+        scheduler.Record([buffer_handle = *buffer, image = blit_image->GetHandle(),
+                          copy](auto cmdbuf, auto& dld) {
+            cmdbuf.copyBufferToImage(buffer_handle, image, vk::ImageLayout::eTransferDstOptimal,
+                                     {copy}, dld);
+        });
+    }
+    map.Release();
+
+    blit_image->Transition(0, 1, 0, 1, vk::PipelineStageFlagBits::eFragmentShader,
+                           vk::AccessFlagBits::eShaderRead,
+                           vk::ImageLayout::eShaderReadOnlyOptimal);
+
+    scheduler.Record([renderpass = *renderpass, framebuffer = *framebuffers[image_index],
+                      descriptor_set = descriptor_sets[image_index], buffer = *buffer,
+                      size = swapchain.GetSize(), pipeline = *pipeline,
+                      layout = *pipeline_layout](auto cmdbuf, auto& dld) {
+        const vk::ClearValue clear_color{std::array{0.0f, 0.0f, 0.0f, 1.0f}};
+        const vk::RenderPassBeginInfo renderpass_bi(renderpass, framebuffer, {{0, 0}, size}, 1,
+                                                    &clear_color);
+
+        cmdbuf.beginRenderPass(renderpass_bi, vk::SubpassContents::eInline, dld);
+        cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline, dld);
+        cmdbuf.setViewport(
+            0,
+            {{0.0f, 0.0f, static_cast<f32>(size.width), static_cast<f32>(size.height), 0.0f, 1.0f}},
+            dld);
+        cmdbuf.setScissor(0, {{{0, 0}, size}}, dld);
+
+        cmdbuf.bindVertexBuffers(0, {buffer}, {offsetof(BufferData, vertices)}, dld);
+        cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, layout, 0, {descriptor_set}, {},
+                                  dld);
+        cmdbuf.draw(4, 1, 0, 0, dld);
+        cmdbuf.endRenderPass(dld);
+    });
+
+    return {scheduler.GetFence(), *semaphores[image_index]};
+}
+
+void VKBlitScreen::CreateStaticResources() {
+    CreateShaders();
+    CreateSemaphores();
+    CreateDescriptorPool();
+    CreateDescriptorSetLayout();
+    CreateDescriptorSets();
+    CreatePipelineLayout();
+    CreateSampler();
+}
+
+void VKBlitScreen::CreateDynamicResources() {
+    CreateRenderPass();
+    CreateFramebuffers();
+    CreateGraphicsPipeline();
+}
+
+void VKBlitScreen::RefreshResources(const Tegra::FramebufferConfig& framebuffer) {
+    if (framebuffer.width == raw_width && framebuffer.height == raw_height && !raw_images.empty()) {
+        return;
+    }
+    raw_width = framebuffer.width;
+    raw_height = framebuffer.height;
+    ReleaseRawImages();
+
+    CreateStagingBuffer(framebuffer);
+    CreateRawImages(framebuffer);
+}
+
+void VKBlitScreen::CreateShaders() {
+    vertex_shader = BuildShader(device, sizeof(blit_vertex_code), blit_vertex_code);
+    fragment_shader = BuildShader(device, sizeof(blit_fragment_code), blit_fragment_code);
+}
+
+void VKBlitScreen::CreateSemaphores() {
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+
+    semaphores.resize(image_count);
+    for (std::size_t i = 0; i < image_count; ++i) {
+        semaphores[i] = dev.createSemaphoreUnique({}, nullptr, dld);
+    }
+}
+
+void VKBlitScreen::CreateDescriptorPool() {
+    const std::array<vk::DescriptorPoolSize, 2> pool_sizes{
+        vk::DescriptorPoolSize{vk::DescriptorType::eUniformBuffer, static_cast<u32>(image_count)},
+        vk::DescriptorPoolSize{vk::DescriptorType::eCombinedImageSampler,
+                               static_cast<u32>(image_count)}};
+    const vk::DescriptorPoolCreateInfo pool_ci(
+        {}, static_cast<u32>(image_count), static_cast<u32>(pool_sizes.size()), pool_sizes.data());
+    const auto dev = device.GetLogical();
+    descriptor_pool = dev.createDescriptorPoolUnique(pool_ci, nullptr, device.GetDispatchLoader());
+}
+
+void VKBlitScreen::CreateRenderPass() {
+    const vk::AttachmentDescription color_attachment(
+        {}, swapchain.GetImageFormat(), vk::SampleCountFlagBits::e1, vk::AttachmentLoadOp::eClear,
+        vk::AttachmentStoreOp::eStore, vk::AttachmentLoadOp::eDontCare,
+        vk::AttachmentStoreOp::eDontCare, vk::ImageLayout::eUndefined,
+        vk::ImageLayout::ePresentSrcKHR);
+
+    const vk::AttachmentReference color_attachment_ref(0, vk::ImageLayout::eColorAttachmentOptimal);
+
+    const vk::SubpassDescription subpass_description({}, vk::PipelineBindPoint::eGraphics, 0,
+                                                     nullptr, 1, &color_attachment_ref, nullptr,
+                                                     nullptr, 0, nullptr);
+
+    const vk::SubpassDependency dependency(
+        VK_SUBPASS_EXTERNAL, 0, vk::PipelineStageFlagBits::eColorAttachmentOutput,
+        vk::PipelineStageFlagBits::eColorAttachmentOutput, {},
+        vk::AccessFlagBits::eColorAttachmentRead | vk::AccessFlagBits::eColorAttachmentWrite, {});
+
+    const vk::RenderPassCreateInfo renderpass_ci({}, 1, &color_attachment, 1, &subpass_description,
+                                                 1, &dependency);
+
+    const auto dev = device.GetLogical();
+    renderpass = dev.createRenderPassUnique(renderpass_ci, nullptr, device.GetDispatchLoader());
+}
+
+void VKBlitScreen::CreateDescriptorSetLayout() {
+    const std::array<vk::DescriptorSetLayoutBinding, 2> layout_bindings{
+        vk::DescriptorSetLayoutBinding(0, vk::DescriptorType::eUniformBuffer, 1,
+                                       vk::ShaderStageFlagBits::eVertex, nullptr),
+        vk::DescriptorSetLayoutBinding(1, vk::DescriptorType::eCombinedImageSampler, 1,
+                                       vk::ShaderStageFlagBits::eFragment, nullptr)};
+    const vk::DescriptorSetLayoutCreateInfo descriptor_layout_ci(
+        {}, static_cast<u32>(layout_bindings.size()), layout_bindings.data());
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    descriptor_set_layout = dev.createDescriptorSetLayoutUnique(descriptor_layout_ci, nullptr, dld);
+}
+
+void VKBlitScreen::CreateDescriptorSets() {
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+
+    descriptor_sets.resize(image_count);
+    for (std::size_t i = 0; i < image_count; ++i) {
+        const vk::DescriptorSetLayout layout = *descriptor_set_layout;
+        const vk::DescriptorSetAllocateInfo descriptor_set_ai(*descriptor_pool, 1, &layout);
+        const vk::Result result =
+            dev.allocateDescriptorSets(&descriptor_set_ai, &descriptor_sets[i], dld);
+        ASSERT(result == vk::Result::eSuccess);
+    }
+}
+
+void VKBlitScreen::CreatePipelineLayout() {
+    const vk::PipelineLayoutCreateInfo pipeline_layout_ci({}, 1, &descriptor_set_layout.get(), 0,
+                                                          nullptr);
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    pipeline_layout = dev.createPipelineLayoutUnique(pipeline_layout_ci, nullptr, dld);
+}
+
+void VKBlitScreen::CreateGraphicsPipeline() {
+    const std::array shader_stages = {
+        vk::PipelineShaderStageCreateInfo({}, vk::ShaderStageFlagBits::eVertex, *vertex_shader,
+                                          "main", nullptr),
+        vk::PipelineShaderStageCreateInfo({}, vk::ShaderStageFlagBits::eFragment, *fragment_shader,
+                                          "main", nullptr)};
+
+    const auto vertex_binding_description = ScreenRectVertex::GetDescription();
+    const auto vertex_attrs_description = ScreenRectVertex::GetAttributes();
+    const vk::PipelineVertexInputStateCreateInfo vertex_input(
+        {}, 1, &vertex_binding_description, static_cast<u32>(vertex_attrs_description.size()),
+        vertex_attrs_description.data());
+
+    const vk::PipelineInputAssemblyStateCreateInfo input_assembly(
+        {}, vk::PrimitiveTopology::eTriangleStrip, false);
+
+    // Set a dummy viewport, it's going to be replaced by dynamic states.
+    const vk::Viewport viewport(0.0f, 0.0f, 1.0f, 1.0f, 0.0f, 1.0f);
+    const vk::Rect2D scissor({0, 0}, {1, 1});
+
+    const vk::PipelineViewportStateCreateInfo viewport_state({}, 1, &viewport, 1, &scissor);
+
+    const vk::PipelineRasterizationStateCreateInfo rasterizer(
+        {}, false, false, vk::PolygonMode::eFill, vk::CullModeFlagBits::eNone,
+        vk::FrontFace::eClockwise, false, 0.0f, 0.0f, 0.0f, 1.0f);
+
+    const vk::PipelineMultisampleStateCreateInfo multisampling({}, vk::SampleCountFlagBits::e1,
+                                                               false, 0.0f, nullptr, false, false);
+
+    const vk::PipelineColorBlendAttachmentState color_blend_attachment(
+        false, vk::BlendFactor::eZero, vk::BlendFactor::eZero, vk::BlendOp::eAdd,
+        vk::BlendFactor::eZero, vk::BlendFactor::eZero, vk::BlendOp::eAdd,
+        vk::ColorComponentFlagBits::eR | vk::ColorComponentFlagBits::eG |
+            vk::ColorComponentFlagBits::eB | vk::ColorComponentFlagBits::eA);
+
+    const vk::PipelineColorBlendStateCreateInfo color_blending(
+        {}, false, vk::LogicOp::eCopy, 1, &color_blend_attachment, {0.0f, 0.0f, 0.0f, 0.0f});
+
+    const std::array<vk::DynamicState, 2> dynamic_states = {vk::DynamicState::eViewport,
+                                                            vk::DynamicState::eScissor};
+
+    const vk::PipelineDynamicStateCreateInfo dynamic_state(
+        {}, static_cast<u32>(dynamic_states.size()), dynamic_states.data());
+
+    const vk::GraphicsPipelineCreateInfo pipeline_ci(
+        {}, static_cast<u32>(shader_stages.size()), shader_stages.data(), &vertex_input,
+        &input_assembly, nullptr, &viewport_state, &rasterizer, &multisampling, nullptr,
+        &color_blending, &dynamic_state, *pipeline_layout, *renderpass, 0, nullptr, 0);
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    pipeline = dev.createGraphicsPipelineUnique({}, pipeline_ci, nullptr, dld);
+}
+
+void VKBlitScreen::CreateSampler() {
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    const vk::SamplerCreateInfo sampler_ci(
+        {}, vk::Filter::eLinear, vk::Filter::eLinear, vk::SamplerMipmapMode::eLinear,
+        vk::SamplerAddressMode::eClampToBorder, vk::SamplerAddressMode::eClampToBorder,
+        vk::SamplerAddressMode::eClampToBorder, 0.0f, false, 0.0f, false, vk::CompareOp::eNever,
+        0.0f, 0.0f, vk::BorderColor::eFloatOpaqueBlack, false);
+    sampler = dev.createSamplerUnique(sampler_ci, nullptr, dld);
+}
+
+void VKBlitScreen::CreateFramebuffers() {
+    const vk::Extent2D size{swapchain.GetSize()};
+    framebuffers.clear();
+    framebuffers.resize(image_count);
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+
+    for (std::size_t i = 0; i < image_count; ++i) {
+        const vk::ImageView image_view{swapchain.GetImageViewIndex(i)};
+        const vk::FramebufferCreateInfo framebuffer_ci({}, *renderpass, 1, &image_view, size.width,
+                                                       size.height, 1);
+        framebuffers[i] = dev.createFramebufferUnique(framebuffer_ci, nullptr, dld);
+    }
+}
+
+void VKBlitScreen::ReleaseRawImages() {
+    for (std::size_t i = 0; i < raw_images.size(); ++i) {
+        watches[i]->Wait();
+    }
+    raw_images.clear();
+    raw_buffer_commits.clear();
+    buffer.reset();
+    buffer_commit.reset();
+}
+
+void VKBlitScreen::CreateStagingBuffer(const Tegra::FramebufferConfig& framebuffer) {
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+
+    const vk::BufferCreateInfo buffer_ci({}, CalculateBufferSize(framebuffer),
+                                         vk::BufferUsageFlagBits::eTransferSrc |
+                                             vk::BufferUsageFlagBits::eVertexBuffer |
+                                             vk::BufferUsageFlagBits::eUniformBuffer,
+                                         vk::SharingMode::eExclusive, 0, nullptr);
+    buffer = dev.createBufferUnique(buffer_ci, nullptr, dld);
+    buffer_commit = memory_manager.Commit(*buffer, true);
+}
+
+void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) {
+    raw_images.resize(image_count);
+    raw_buffer_commits.resize(image_count);
+
+    const auto format = GetFormat(framebuffer);
+    for (std::size_t i = 0; i < image_count; ++i) {
+        const vk::ImageCreateInfo image_ci(
+            {}, vk::ImageType::e2D, format, {framebuffer.width, framebuffer.height, 1}, 1, 1,
+            vk::SampleCountFlagBits::e1, vk::ImageTiling::eOptimal,
+            vk::ImageUsageFlagBits::eTransferDst | vk::ImageUsageFlagBits::eSampled,
+            vk::SharingMode::eExclusive, 0, nullptr, vk::ImageLayout::eUndefined);
+
+        raw_images[i] =
+            std::make_unique<VKImage>(device, scheduler, image_ci, vk::ImageAspectFlagBits::eColor);
+        raw_buffer_commits[i] = memory_manager.Commit(raw_images[i]->GetHandle(), false);
+    }
+}
+
+void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, vk::ImageView image_view) const {
+    const vk::DescriptorSet descriptor_set = descriptor_sets[image_index];
+
+    const vk::DescriptorBufferInfo buffer_info(*buffer, offsetof(BufferData, uniform),
+                                               sizeof(BufferData::uniform));
+    const vk::WriteDescriptorSet ubo_write(descriptor_set, 0, 0, 1,
+                                           vk::DescriptorType::eUniformBuffer, nullptr,
+                                           &buffer_info, nullptr);
+
+    const vk::DescriptorImageInfo image_info(*sampler, image_view,
+                                             vk::ImageLayout::eShaderReadOnlyOptimal);
+    const vk::WriteDescriptorSet sampler_write(descriptor_set, 1, 0, 1,
+                                               vk::DescriptorType::eCombinedImageSampler,
+                                               &image_info, nullptr, nullptr);
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    dev.updateDescriptorSets({ubo_write, sampler_write}, {}, dld);
+}
+
+void VKBlitScreen::SetUniformData(BufferData& data,
+                                  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,
+                                 const Tegra::FramebufferConfig& framebuffer) 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 Tegra::FramebufferConfig::TransformFlags::Unset:
+        break;
+    case Tegra::FramebufferConfig::TransformFlags::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;
+    }
+
+    UNIMPLEMENTED_IF(framebuffer_crop_rect.top != 0);
+    UNIMPLEMENTED_IF(framebuffer_crop_rect.left != 0);
+
+    // Scale the output by the crop width/height. This is commonly used with 1280x720 rendering
+    // (e.g. handheld mode) on a 1920x1080 framebuffer.
+    f32 scale_u = 1.0f;
+    f32 scale_v = 1.0f;
+    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);
+    }
+
+    const auto& screen = render_window.GetFramebufferLayout().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 * scale_v);
+    data.vertices[1] = ScreenRectVertex(x + w, y, texcoords.bottom * scale_u, left * scale_v);
+    data.vertices[2] = ScreenRectVertex(x, y + h, texcoords.top * scale_u, right * scale_v);
+    data.vertices[3] = ScreenRectVertex(x + w, y + h, texcoords.bottom * scale_u, right * scale_v);
+}
+
+u64 VKBlitScreen::CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const {
+    return sizeof(BufferData) + GetSizeInBytes(framebuffer) * image_count;
+}
+
+u64 VKBlitScreen::GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer,
+                                    std::size_t image_index) const {
+    constexpr auto first_image_offset = static_cast<u64>(sizeof(BufferData));
+    return first_image_offset + GetSizeInBytes(framebuffer) * image_index;
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_blit_screen.h

@@ -0,0 +1,119 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <memory>
+#include <tuple>
+
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_memory_manager.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+
+namespace Core {
+class System;
+}
+
+namespace Core::Frontend {
+class EmuWindow;
+}
+
+namespace Tegra {
+struct FramebufferConfig;
+}
+
+namespace VideoCore {
+class RasterizerInterface;
+}
+
+namespace Vulkan {
+
+struct ScreenInfo;
+class RasterizerVulkan;
+class VKDevice;
+class VKFence;
+class VKImage;
+class VKScheduler;
+class VKSwapchain;
+
+class VKBlitScreen final {
+public:
+    explicit VKBlitScreen(Core::System& system, Core::Frontend::EmuWindow& render_window,
+                          VideoCore::RasterizerInterface& rasterizer, const VKDevice& device,
+                          VKResourceManager& resource_manager, VKMemoryManager& memory_manager,
+                          VKSwapchain& swapchain, VKScheduler& scheduler,
+                          const VKScreenInfo& screen_info);
+    ~VKBlitScreen();
+
+    void Recreate();
+
+    std::tuple<VKFence&, vk::Semaphore> Draw(const Tegra::FramebufferConfig& framebuffer,
+                                             bool use_accelerated);
+
+private:
+    struct BufferData;
+
+    void CreateStaticResources();
+    void CreateShaders();
+    void CreateSemaphores();
+    void CreateDescriptorPool();
+    void CreateRenderPass();
+    void CreateDescriptorSetLayout();
+    void CreateDescriptorSets();
+    void CreatePipelineLayout();
+    void CreateGraphicsPipeline();
+    void CreateSampler();
+
+    void CreateDynamicResources();
+    void CreateFramebuffers();
+
+    void RefreshResources(const Tegra::FramebufferConfig& framebuffer);
+    void ReleaseRawImages();
+    void CreateStagingBuffer(const Tegra::FramebufferConfig& framebuffer);
+    void CreateRawImages(const Tegra::FramebufferConfig& framebuffer);
+
+    void UpdateDescriptorSet(std::size_t image_index, vk::ImageView image_view) const;
+    void SetUniformData(BufferData& data, const Tegra::FramebufferConfig& framebuffer) const;
+    void SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer) const;
+
+    u64 CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const;
+    u64 GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer,
+                          std::size_t image_index) const;
+
+    Core::System& system;
+    Core::Frontend::EmuWindow& render_window;
+    VideoCore::RasterizerInterface& rasterizer;
+    const VKDevice& device;
+    VKResourceManager& resource_manager;
+    VKMemoryManager& memory_manager;
+    VKSwapchain& swapchain;
+    VKScheduler& scheduler;
+    const std::size_t image_count;
+    const VKScreenInfo& screen_info;
+
+    UniqueShaderModule vertex_shader;
+    UniqueShaderModule fragment_shader;
+    UniqueDescriptorPool descriptor_pool;
+    UniqueDescriptorSetLayout descriptor_set_layout;
+    UniquePipelineLayout pipeline_layout;
+    UniquePipeline pipeline;
+    UniqueRenderPass renderpass;
+    std::vector<UniqueFramebuffer> framebuffers;
+    std::vector<vk::DescriptorSet> descriptor_sets;
+    UniqueSampler sampler;
+
+    UniqueBuffer buffer;
+    VKMemoryCommit buffer_commit;
+
+    std::vector<std::unique_ptr<VKFenceWatch>> watches;
+
+    std::vector<UniqueSemaphore> semaphores;
+    std::vector<std::unique_ptr<VKImage>> raw_images;
+    std::vector<VKMemoryCommit> raw_buffer_commits;
+    u32 raw_width = 0;
+    u32 raw_height = 0;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -2,124 +2,145 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <algorithm>
 #include <cstring>
 #include <memory>
 #include <optional>
 #include <tuple>
 
-#include "common/alignment.h"
 #include "common/assert.h"
-#include "core/memory.h"
-#include "video_core/memory_manager.h"
+#include "common/bit_util.h"
+#include "core/core.h"
 #include "video_core/renderer_vulkan/declarations.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
+#include "video_core/renderer_vulkan/vk_device.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_stream_buffer.h"
 
 namespace Vulkan {
 
-CachedBufferEntry::CachedBufferEntry(VAddr cpu_addr, std::size_t size, u64 offset,
-                                     std::size_t alignment, u8* host_ptr)
-    : RasterizerCacheObject{host_ptr}, cpu_addr{cpu_addr}, size{size}, offset{offset},
-      alignment{alignment} {}
+namespace {
 
-VKBufferCache::VKBufferCache(Tegra::MemoryManager& tegra_memory_manager,
-                             Memory::Memory& cpu_memory_,
-                             VideoCore::RasterizerInterface& rasterizer, const VKDevice& device,
-                             VKMemoryManager& memory_manager, VKScheduler& scheduler, u64 size)
-    : RasterizerCache{rasterizer}, tegra_memory_manager{tegra_memory_manager}, cpu_memory{
-                                                                                   cpu_memory_} {
-    const auto usage = vk::BufferUsageFlagBits::eVertexBuffer |
-                       vk::BufferUsageFlagBits::eIndexBuffer |
-                       vk::BufferUsageFlagBits::eUniformBuffer;
-    const auto access = vk::AccessFlagBits::eVertexAttributeRead | vk::AccessFlagBits::eIndexRead |
-                        vk::AccessFlagBits::eUniformRead;
-    stream_buffer =
-        std::make_unique<VKStreamBuffer>(device, memory_manager, scheduler, size, usage, access,
-                                         vk::PipelineStageFlagBits::eAllCommands);
-    buffer_handle = stream_buffer->GetBuffer();
+const auto BufferUsage =
+    vk::BufferUsageFlagBits::eVertexBuffer | vk::BufferUsageFlagBits::eIndexBuffer |
+    vk::BufferUsageFlagBits::eUniformBuffer | vk::BufferUsageFlagBits::eStorageBuffer;
+
+const auto UploadPipelineStage =
+    vk::PipelineStageFlagBits::eTransfer | vk::PipelineStageFlagBits::eVertexInput |
+    vk::PipelineStageFlagBits::eVertexShader | vk::PipelineStageFlagBits::eFragmentShader |
+    vk::PipelineStageFlagBits::eComputeShader;
+
+const auto UploadAccessBarriers =
+    vk::AccessFlagBits::eTransferRead | vk::AccessFlagBits::eShaderRead |
+    vk::AccessFlagBits::eUniformRead | vk::AccessFlagBits::eVertexAttributeRead |
+    vk::AccessFlagBits::eIndexRead;
+
+auto CreateStreamBuffer(const VKDevice& device, VKScheduler& scheduler) {
+    return std::make_unique<VKStreamBuffer>(device, scheduler, BufferUsage);
 }
 
+} // Anonymous namespace
+
+CachedBufferBlock::CachedBufferBlock(const VKDevice& device, VKMemoryManager& memory_manager,
+                                     CacheAddr cache_addr, std::size_t size)
+    : VideoCommon::BufferBlock{cache_addr, size} {
+    const vk::BufferCreateInfo buffer_ci({}, static_cast<vk::DeviceSize>(size),
+                                         BufferUsage | vk::BufferUsageFlagBits::eTransferSrc |
+                                             vk::BufferUsageFlagBits::eTransferDst,
+                                         vk::SharingMode::eExclusive, 0, nullptr);
+
+    const auto& dld{device.GetDispatchLoader()};
+    const auto dev{device.GetLogical()};
+    buffer.handle = dev.createBufferUnique(buffer_ci, nullptr, dld);
+    buffer.commit = memory_manager.Commit(*buffer.handle, false);
+}
+
+CachedBufferBlock::~CachedBufferBlock() = default;
+
+VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
+                             const VKDevice& device, VKMemoryManager& memory_manager,
+                             VKScheduler& scheduler, VKStagingBufferPool& staging_pool)
+    : VideoCommon::BufferCache<Buffer, vk::Buffer, VKStreamBuffer>{rasterizer, system,
+                                                                   CreateStreamBuffer(device,
+                                                                                      scheduler)},
+      device{device}, memory_manager{memory_manager}, scheduler{scheduler}, staging_pool{
+                                                                                staging_pool} {}
+
 VKBufferCache::~VKBufferCache() = default;
 
-u64 VKBufferCache::UploadMemory(GPUVAddr gpu_addr, std::size_t size, u64 alignment, bool cache) {
-    const auto cpu_addr{tegra_memory_manager.GpuToCpuAddress(gpu_addr)};
-    ASSERT_MSG(cpu_addr, "Invalid GPU address");
-
-    // Cache management is a big overhead, so only cache entries with a given size.
-    // TODO: Figure out which size is the best for given games.
-    cache &= size >= 2048;
-
-    u8* const host_ptr{cpu_memory.GetPointer(*cpu_addr)};
-    if (cache) {
-        const auto entry = TryGet(host_ptr);
-        if (entry) {
-            if (entry->GetSize() >= size && entry->GetAlignment() == alignment) {
-                return entry->GetOffset();
-            }
-            Unregister(entry);
-        }
-    }
-
-    AlignBuffer(alignment);
-    const u64 uploaded_offset = buffer_offset;
-
-    if (host_ptr == nullptr) {
-        return uploaded_offset;
-    }
-
-    std::memcpy(buffer_ptr, host_ptr, size);
-    buffer_ptr += size;
-    buffer_offset += size;
-
-    if (cache) {
-        auto entry = std::make_shared<CachedBufferEntry>(*cpu_addr, size, uploaded_offset,
-                                                         alignment, host_ptr);
-        Register(entry);
-    }
-
-    return uploaded_offset;
+Buffer VKBufferCache::CreateBlock(CacheAddr cache_addr, std::size_t size) {
+    return std::make_shared<CachedBufferBlock>(device, memory_manager, cache_addr, size);
 }
 
-u64 VKBufferCache::UploadHostMemory(const u8* raw_pointer, std::size_t size, u64 alignment) {
-    AlignBuffer(alignment);
-    std::memcpy(buffer_ptr, raw_pointer, size);
-    const u64 uploaded_offset = buffer_offset;
-
-    buffer_ptr += size;
-    buffer_offset += size;
-    return uploaded_offset;
+const vk::Buffer* VKBufferCache::ToHandle(const Buffer& buffer) {
+    return buffer->GetHandle();
 }
 
-std::tuple<u8*, u64> VKBufferCache::ReserveMemory(std::size_t size, u64 alignment) {
-    AlignBuffer(alignment);
-    u8* const uploaded_ptr = buffer_ptr;
-    const u64 uploaded_offset = buffer_offset;
-
-    buffer_ptr += size;
-    buffer_offset += size;
-    return {uploaded_ptr, uploaded_offset};
+const vk::Buffer* VKBufferCache::GetEmptyBuffer(std::size_t size) {
+    size = std::max(size, std::size_t(4));
+    const auto& empty = staging_pool.GetUnusedBuffer(size, false);
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([size, buffer = *empty.handle](vk::CommandBuffer cmdbuf, auto& dld) {
+        cmdbuf.fillBuffer(buffer, 0, size, 0, dld);
+    });
+    return &*empty.handle;
 }
 
-void VKBufferCache::Reserve(std::size_t max_size) {
-    bool invalidate;
-    std::tie(buffer_ptr, buffer_offset_base, invalidate) = stream_buffer->Reserve(max_size);
-    buffer_offset = buffer_offset_base;
+void VKBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                                    const u8* data) {
+    const auto& staging = staging_pool.GetUnusedBuffer(size, true);
+    std::memcpy(staging.commit->Map(size), data, size);
 
-    if (invalidate) {
-        InvalidateAll();
-    }
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([staging = *staging.handle, buffer = *buffer->GetHandle(), offset,
+                      size](auto cmdbuf, auto& dld) {
+        cmdbuf.copyBuffer(staging, buffer, {{0, offset, size}}, dld);
+        cmdbuf.pipelineBarrier(
+            vk::PipelineStageFlagBits::eTransfer, UploadPipelineStage, {}, {},
+            {vk::BufferMemoryBarrier(vk::AccessFlagBits::eTransferWrite, UploadAccessBarriers,
+                                     VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED, buffer,
+                                     offset, size)},
+            {}, dld);
+    });
 }
 
-void VKBufferCache::Send() {
-    stream_buffer->Send(buffer_offset - buffer_offset_base);
+void VKBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                                      u8* data) {
+    const auto& staging = staging_pool.GetUnusedBuffer(size, true);
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([staging = *staging.handle, buffer = *buffer->GetHandle(), offset,
+                      size](auto cmdbuf, auto& dld) {
+        cmdbuf.pipelineBarrier(
+            vk::PipelineStageFlagBits::eVertexShader | vk::PipelineStageFlagBits::eFragmentShader |
+                vk::PipelineStageFlagBits::eComputeShader,
+            vk::PipelineStageFlagBits::eTransfer, {}, {},
+            {vk::BufferMemoryBarrier(vk::AccessFlagBits::eShaderWrite,
+                                     vk::AccessFlagBits::eTransferRead, VK_QUEUE_FAMILY_IGNORED,
+                                     VK_QUEUE_FAMILY_IGNORED, buffer, offset, size)},
+            {}, dld);
+        cmdbuf.copyBuffer(buffer, staging, {{offset, 0, size}}, dld);
+    });
+    scheduler.Finish();
+
+    std::memcpy(data, staging.commit->Map(size), size);
 }
 
-void VKBufferCache::AlignBuffer(std::size_t alignment) {
-    // Align the offset, not the mapped pointer
-    const u64 offset_aligned = Common::AlignUp(buffer_offset, alignment);
-    buffer_ptr += offset_aligned - buffer_offset;
-    buffer_offset = offset_aligned;
+void VKBufferCache::CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
+                              std::size_t dst_offset, std::size_t size) {
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([src_buffer = *src->GetHandle(), dst_buffer = *dst->GetHandle(), src_offset,
+                      dst_offset, size](auto cmdbuf, auto& dld) {
+        cmdbuf.copyBuffer(src_buffer, dst_buffer, {{src_offset, dst_offset, size}}, dld);
+        cmdbuf.pipelineBarrier(
+            vk::PipelineStageFlagBits::eTransfer, UploadPipelineStage, {}, {},
+            {vk::BufferMemoryBarrier(vk::AccessFlagBits::eTransferRead,
+                                     vk::AccessFlagBits::eShaderWrite, VK_QUEUE_FAMILY_IGNORED,
+                                     VK_QUEUE_FAMILY_IGNORED, src_buffer, src_offset, size),
+             vk::BufferMemoryBarrier(vk::AccessFlagBits::eTransferWrite, UploadAccessBarriers,
+                                     VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED, dst_buffer,
+                                     dst_offset, size)},
+            {}, dld);
+    });
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -5,105 +5,74 @@
 #pragma once
 
 #include <memory>
-#include <tuple>
+#include <unordered_map>
+#include <vector>
 
 #include "common/common_types.h"
-#include "video_core/gpu.h"
+#include "video_core/buffer_cache/buffer_cache.h"
 #include "video_core/rasterizer_cache.h"
 #include "video_core/renderer_vulkan/declarations.h"
-#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_memory_manager.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
+#include "video_core/renderer_vulkan/vk_stream_buffer.h"
 
-namespace Memory {
-class Memory;
-}
-
-namespace Tegra {
-class MemoryManager;
+namespace Core {
+class System;
 }
 
 namespace Vulkan {
 
 class VKDevice;
-class VKFence;
 class VKMemoryManager;
-class VKStreamBuffer;
+class VKScheduler;
 
-class CachedBufferEntry final : public RasterizerCacheObject {
+class CachedBufferBlock final : public VideoCommon::BufferBlock {
 public:
-    explicit CachedBufferEntry(VAddr cpu_addr, std::size_t size, u64 offset, std::size_t alignment,
-                               u8* host_ptr);
+    explicit CachedBufferBlock(const VKDevice& device, VKMemoryManager& memory_manager,
+                               CacheAddr cache_addr, std::size_t size);
+    ~CachedBufferBlock();
 
-    VAddr GetCpuAddr() const override {
-        return cpu_addr;
-    }
-
-    std::size_t GetSizeInBytes() const override {
-        return size;
-    }
-
-    std::size_t GetSize() const {
-        return size;
-    }
-
-    u64 GetOffset() const {
-        return offset;
-    }
-
-    std::size_t GetAlignment() const {
-        return alignment;
+    const vk::Buffer* GetHandle() const {
+        return &*buffer.handle;
     }
 
 private:
-    VAddr cpu_addr{};
-    std::size_t size{};
-    u64 offset{};
-    std::size_t alignment{};
+    VKBuffer buffer;
 };
 
-class VKBufferCache final : public RasterizerCache<std::shared_ptr<CachedBufferEntry>> {
+using Buffer = std::shared_ptr<CachedBufferBlock>;
+
+class VKBufferCache final : public VideoCommon::BufferCache<Buffer, vk::Buffer, VKStreamBuffer> {
 public:
-    explicit VKBufferCache(Tegra::MemoryManager& tegra_memory_manager, Memory::Memory& cpu_memory_,
-                           VideoCore::RasterizerInterface& rasterizer, const VKDevice& device,
-                           VKMemoryManager& memory_manager, VKScheduler& scheduler, u64 size);
+    explicit VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
+                           const VKDevice& device, VKMemoryManager& memory_manager,
+                           VKScheduler& scheduler, VKStagingBufferPool& staging_pool);
     ~VKBufferCache();
 
-    /// Uploads data from a guest GPU address. Returns host's buffer offset where it's been
-    /// allocated.
-    u64 UploadMemory(GPUVAddr gpu_addr, std::size_t size, u64 alignment = 4, bool cache = true);
-
-    /// Uploads from a host memory. Returns host's buffer offset where it's been allocated.
-    u64 UploadHostMemory(const u8* raw_pointer, std::size_t size, u64 alignment = 4);
-
-    /// Reserves memory to be used by host's CPU. Returns mapped address and offset.
-    std::tuple<u8*, u64> ReserveMemory(std::size_t size, u64 alignment = 4);
-
-    /// Reserves a region of memory to be used in subsequent upload/reserve operations.
-    void Reserve(std::size_t max_size);
-
-    /// Ensures that the set data is sent to the device.
-    void Send();
-
-    /// Returns the buffer cache handle.
-    vk::Buffer GetBuffer() const {
-        return buffer_handle;
-    }
+    const vk::Buffer* GetEmptyBuffer(std::size_t size) override;
 
 protected:
-    // We do not have to flush this cache as things in it are never modified by us.
-    void FlushObjectInner(const std::shared_ptr<CachedBufferEntry>& object) override {}
+    void WriteBarrier() override {}
+
+    Buffer CreateBlock(CacheAddr cache_addr, std::size_t size) override;
+
+    const vk::Buffer* ToHandle(const Buffer& buffer) override;
+
+    void UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                         const u8* data) override;
+
+    void DownloadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
+                           u8* data) override;
+
+    void CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
+                   std::size_t dst_offset, std::size_t size) override;
 
 private:
-    void AlignBuffer(std::size_t alignment);
-
-    Tegra::MemoryManager& tegra_memory_manager;
-    Memory::Memory& cpu_memory;
-
-    std::unique_ptr<VKStreamBuffer> stream_buffer;
-    vk::Buffer buffer_handle;
-
-    u8* buffer_ptr = nullptr;
-    u64 buffer_offset = 0;
-    u64 buffer_offset_base = 0;
+    const VKDevice& device;
+    VKMemoryManager& memory_manager;
+    VKScheduler& scheduler;
+    VKStagingBufferPool& staging_pool;
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -0,0 +1,339 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <cstring>
+#include <memory>
+#include <optional>
+#include <utility>
+#include <vector>
+#include "common/alignment.h"
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_compute_pass.h"
+#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
+#include "video_core/renderer_vulkan/vk_update_descriptor.h"
+
+namespace Vulkan {
+
+namespace {
+
+// Quad array SPIR-V module. Generated from the "shaders/" directory, read the instructions there.
+constexpr u8 quad_array[] = {
+    0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x54, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
+    0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x0f, 0x00, 0x06, 0x00, 0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e,
+    0x00, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00,
+    0x11, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x47, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
+    0x47, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
+    0x48, 0x00, 0x05, 0x00, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x14, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
+    0x47, 0x00, 0x04, 0x00, 0x16, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x47, 0x00, 0x04, 0x00, 0x16, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x48, 0x00, 0x05, 0x00, 0x29, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x29, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x47, 0x00, 0x04, 0x00, 0x4a, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
+    0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00, 0x03, 0x00, 0x00, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00, 0x13, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x1e, 0x00, 0x03, 0x00, 0x14, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
+    0x15, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
+    0x15, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
+    0x18, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x14, 0x00, 0x02, 0x00,
+    0x1b, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x29, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x20, 0x00, 0x04, 0x00, 0x2a, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00,
+    0x3b, 0x00, 0x04, 0x00, 0x2a, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
+    0x2b, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x20, 0x00, 0x04, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
+    0x1c, 0x00, 0x04, 0x00, 0x34, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00,
+    0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x37, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
+    0x2c, 0x00, 0x09, 0x00, 0x34, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00,
+    0x35, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00,
+    0x37, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x3a, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00,
+    0x34, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x44, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x47, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00,
+    0x00, 0x04, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00, 0x4a, 0x00, 0x00, 0x00,
+    0x49, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x35, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
+    0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x3a, 0x00, 0x00, 0x00,
+    0x3b, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x4c, 0x00, 0x00, 0x00,
+    0xf8, 0x00, 0x02, 0x00, 0x4c, 0x00, 0x00, 0x00, 0xf6, 0x00, 0x04, 0x00, 0x4b, 0x00, 0x00, 0x00,
+    0x4e, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x4d, 0x00, 0x00, 0x00,
+    0xf8, 0x00, 0x02, 0x00, 0x4d, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0d, 0x00, 0x00, 0x00,
+    0x0e, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x12, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00,
+    0x44, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x17, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
+    0x17, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00,
+    0x19, 0x00, 0x00, 0x00, 0xae, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
+    0x12, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0xf7, 0x00, 0x03, 0x00, 0x1e, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
+    0x1e, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00,
+    0x4b, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1e, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00,
+    0x21, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x21, 0x00, 0x00, 0x00, 0xf5, 0x00, 0x07, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00,
+    0x48, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0xb0, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00,
+    0x27, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0xf6, 0x00, 0x04, 0x00,
+    0x23, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00,
+    0x27, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00,
+    0x22, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00,
+    0x2b, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x2f, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x32, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x33, 0x00, 0x00, 0x00, 0x2f, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00,
+    0x3e, 0x00, 0x03, 0x00, 0x3b, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00,
+    0x07, 0x00, 0x00, 0x00, 0x3c, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00,
+    0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x00, 0x00, 0x3c, 0x00, 0x00, 0x00,
+    0x80, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x00, 0x00, 0x33, 0x00, 0x00, 0x00,
+    0x3d, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00,
+    0x12, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x44, 0x00, 0x00, 0x00,
+    0x45, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00,
+    0x3e, 0x00, 0x03, 0x00, 0x45, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x00, 0x53, 0x00, 0x00, 0x00, 0x47, 0x00, 0x00, 0x00,
+    0xf9, 0x00, 0x02, 0x00, 0x21, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0xf9, 0x00, 0x02, 0x00, 0x4b, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x4e, 0x00, 0x00, 0x00,
+    0xf9, 0x00, 0x02, 0x00, 0x4c, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x4b, 0x00, 0x00, 0x00,
+    0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00};
+
+// Uint8 SPIR-V module. Generated from the "shaders/" directory.
+constexpr u8 uint8_pass[] = {
+    0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x2f, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00,
+    0x51, 0x11, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x61, 0x11, 0x00, 0x00, 0x0a, 0x00, 0x07, 0x00,
+    0x53, 0x50, 0x56, 0x5f, 0x4b, 0x48, 0x52, 0x5f, 0x31, 0x36, 0x62, 0x69, 0x74, 0x5f, 0x73, 0x74,
+    0x6f, 0x72, 0x61, 0x67, 0x65, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x07, 0x00, 0x53, 0x50, 0x56, 0x5f,
+    0x4b, 0x48, 0x52, 0x5f, 0x38, 0x62, 0x69, 0x74, 0x5f, 0x73, 0x74, 0x6f, 0x72, 0x61, 0x67, 0x65,
+    0x00, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c,
+    0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30, 0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x06, 0x00, 0x05, 0x00, 0x00, 0x00,
+    0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e, 0x00, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
+    0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
+    0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x12, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x48, 0x00, 0x04, 0x00, 0x13, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x13, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00,
+    0x13, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x15, 0x00, 0x00, 0x00,
+    0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x15, 0x00, 0x00, 0x00,
+    0x21, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x1f, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x48, 0x00, 0x04, 0x00, 0x20, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x20, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00,
+    0x20, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
+    0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
+    0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x2e, 0x00, 0x00, 0x00,
+    0x0b, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x21, 0x00, 0x03, 0x00, 0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
+    0x07, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
+    0x0a, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00,
+    0x0a, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
+    0x0d, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
+    0x11, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
+    0x12, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x13, 0x00, 0x00, 0x00,
+    0x12, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x14, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x13, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x14, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x14, 0x00, 0x02, 0x00, 0x1a, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00,
+    0x1e, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
+    0x1f, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x20, 0x00, 0x00, 0x00,
+    0x1f, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x20, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x21, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x26, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x11, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x2a, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x1e, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x00, 0x00,
+    0x00, 0x04, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2d, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00,
+    0x2c, 0x00, 0x00, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x2d, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
+    0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00,
+    0x08, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0d, 0x00, 0x00, 0x00,
+    0x0e, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00,
+    0x08, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x10, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x44, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x16, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
+    0x17, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0xb0, 0x00, 0x05, 0x00,
+    0x1a, 0x00, 0x00, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
+    0xf7, 0x00, 0x03, 0x00, 0x1d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00,
+    0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00,
+    0x1c, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
+    0x08, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
+    0x08, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x26, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
+    0x15, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
+    0x11, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x71, 0x00, 0x04, 0x00,
+    0x1e, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00,
+    0x2a, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0x24, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00,
+    0xf9, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00,
+    0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00};
+
+} // Anonymous namespace
+
+VKComputePass::VKComputePass(const VKDevice& device, VKDescriptorPool& descriptor_pool,
+                             const std::vector<vk::DescriptorSetLayoutBinding>& bindings,
+                             const std::vector<vk::DescriptorUpdateTemplateEntry>& templates,
+                             const std::vector<vk::PushConstantRange> push_constants,
+                             std::size_t code_size, const u8* code) {
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+
+    const vk::DescriptorSetLayoutCreateInfo descriptor_layout_ci(
+        {}, static_cast<u32>(bindings.size()), bindings.data());
+    descriptor_set_layout = dev.createDescriptorSetLayoutUnique(descriptor_layout_ci, nullptr, dld);
+
+    const vk::PipelineLayoutCreateInfo pipeline_layout_ci({}, 1, &*descriptor_set_layout,
+                                                          static_cast<u32>(push_constants.size()),
+                                                          push_constants.data());
+    layout = dev.createPipelineLayoutUnique(pipeline_layout_ci, nullptr, dld);
+
+    if (!templates.empty()) {
+        const vk::DescriptorUpdateTemplateCreateInfo template_ci(
+            {}, static_cast<u32>(templates.size()), templates.data(),
+            vk::DescriptorUpdateTemplateType::eDescriptorSet, *descriptor_set_layout,
+            vk::PipelineBindPoint::eGraphics, *layout, 0);
+        descriptor_template = dev.createDescriptorUpdateTemplateUnique(template_ci, nullptr, dld);
+
+        descriptor_allocator.emplace(descriptor_pool, *descriptor_set_layout);
+    }
+
+    auto code_copy = std::make_unique<u32[]>(code_size / sizeof(u32) + 1);
+    std::memcpy(code_copy.get(), code, code_size);
+    const vk::ShaderModuleCreateInfo module_ci({}, code_size, code_copy.get());
+    module = dev.createShaderModuleUnique(module_ci, nullptr, dld);
+
+    const vk::PipelineShaderStageCreateInfo stage_ci({}, vk::ShaderStageFlagBits::eCompute, *module,
+                                                     "main", nullptr);
+
+    const vk::ComputePipelineCreateInfo pipeline_ci({}, stage_ci, *layout, nullptr, 0);
+    pipeline = dev.createComputePipelineUnique(nullptr, pipeline_ci, nullptr, dld);
+}
+
+VKComputePass::~VKComputePass() = default;
+
+vk::DescriptorSet VKComputePass::CommitDescriptorSet(
+    VKUpdateDescriptorQueue& update_descriptor_queue, VKFence& fence) {
+    if (!descriptor_template) {
+        return {};
+    }
+    const auto set = descriptor_allocator->Commit(fence);
+    update_descriptor_queue.Send(*descriptor_template, set);
+    return set;
+}
+
+QuadArrayPass::QuadArrayPass(const VKDevice& device, VKScheduler& scheduler,
+                             VKDescriptorPool& descriptor_pool,
+                             VKStagingBufferPool& staging_buffer_pool,
+                             VKUpdateDescriptorQueue& update_descriptor_queue)
+    : VKComputePass(device, descriptor_pool,
+                    {vk::DescriptorSetLayoutBinding(0, vk::DescriptorType::eStorageBuffer, 1,
+                                                    vk::ShaderStageFlagBits::eCompute, nullptr)},
+                    {vk::DescriptorUpdateTemplateEntry(0, 0, 1, vk::DescriptorType::eStorageBuffer,
+                                                       0, sizeof(DescriptorUpdateEntry))},
+                    {vk::PushConstantRange(vk::ShaderStageFlagBits::eCompute, 0, sizeof(u32))},
+                    std::size(quad_array), quad_array),
+      scheduler{scheduler}, staging_buffer_pool{staging_buffer_pool},
+      update_descriptor_queue{update_descriptor_queue} {}
+
+QuadArrayPass::~QuadArrayPass() = default;
+
+std::pair<const vk::Buffer&, vk::DeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 first) {
+    const u32 num_triangle_vertices = num_vertices * 6 / 4;
+    const std::size_t staging_size = num_triangle_vertices * sizeof(u32);
+    auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
+
+    update_descriptor_queue.Acquire();
+    update_descriptor_queue.AddBuffer(&*buffer.handle, 0, staging_size);
+    const auto set = CommitDescriptorSet(update_descriptor_queue, scheduler.GetFence());
+
+    scheduler.RequestOutsideRenderPassOperationContext();
+
+    ASSERT(num_vertices % 4 == 0);
+    const u32 num_quads = num_vertices / 4;
+    scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, num_quads,
+                      first, set](auto cmdbuf, auto& dld) {
+        constexpr u32 dispatch_size = 1024;
+        cmdbuf.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline, dld);
+        cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eCompute, layout, 0, {set}, {}, dld);
+        cmdbuf.pushConstants(layout, vk::ShaderStageFlagBits::eCompute, 0, sizeof(first), &first,
+                             dld);
+        cmdbuf.dispatch(Common::AlignUp(num_quads, dispatch_size) / dispatch_size, 1, 1, dld);
+
+        const vk::BufferMemoryBarrier barrier(
+            vk::AccessFlagBits::eShaderWrite, vk::AccessFlagBits::eVertexAttributeRead,
+            VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED, buffer, 0,
+            static_cast<vk::DeviceSize>(num_quads) * 6 * sizeof(u32));
+        cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eComputeShader,
+                               vk::PipelineStageFlagBits::eVertexInput, {}, {}, {barrier}, {}, dld);
+    });
+    return {*buffer.handle, 0};
+}
+
+Uint8Pass::Uint8Pass(const VKDevice& device, VKScheduler& scheduler,
+                     VKDescriptorPool& descriptor_pool, VKStagingBufferPool& staging_buffer_pool,
+                     VKUpdateDescriptorQueue& update_descriptor_queue)
+    : VKComputePass(device, descriptor_pool,
+                    {vk::DescriptorSetLayoutBinding(0, vk::DescriptorType::eStorageBuffer, 1,
+                                                    vk::ShaderStageFlagBits::eCompute, nullptr),
+                     vk::DescriptorSetLayoutBinding(1, vk::DescriptorType::eStorageBuffer, 1,
+                                                    vk::ShaderStageFlagBits::eCompute, nullptr)},
+                    {vk::DescriptorUpdateTemplateEntry(0, 0, 2, vk::DescriptorType::eStorageBuffer,
+                                                       0, sizeof(DescriptorUpdateEntry))},
+                    {}, std::size(uint8_pass), uint8_pass),
+      scheduler{scheduler}, staging_buffer_pool{staging_buffer_pool},
+      update_descriptor_queue{update_descriptor_queue} {}
+
+Uint8Pass::~Uint8Pass() = default;
+
+std::pair<const vk::Buffer*, u64> Uint8Pass::Assemble(u32 num_vertices, vk::Buffer src_buffer,
+                                                      u64 src_offset) {
+    const auto staging_size = static_cast<u32>(num_vertices * sizeof(u16));
+    auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
+
+    update_descriptor_queue.Acquire();
+    update_descriptor_queue.AddBuffer(&src_buffer, src_offset, num_vertices);
+    update_descriptor_queue.AddBuffer(&*buffer.handle, 0, staging_size);
+    const auto set = CommitDescriptorSet(update_descriptor_queue, scheduler.GetFence());
+
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, set,
+                      num_vertices](auto cmdbuf, auto& dld) {
+        constexpr u32 dispatch_size = 1024;
+        cmdbuf.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline, dld);
+        cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eCompute, layout, 0, {set}, {}, dld);
+        cmdbuf.dispatch(Common::AlignUp(num_vertices, dispatch_size) / dispatch_size, 1, 1, dld);
+
+        const vk::BufferMemoryBarrier barrier(
+            vk::AccessFlagBits::eShaderWrite, vk::AccessFlagBits::eVertexAttributeRead,
+            VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED, buffer, 0,
+            static_cast<vk::DeviceSize>(num_vertices) * sizeof(u16));
+        cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eComputeShader,
+                               vk::PipelineStageFlagBits::eVertexInput, {}, {}, {barrier}, {}, dld);
+    });
+    return {&*buffer.handle, 0};
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pass.h

@@ -0,0 +1,77 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <optional>
+#include <utility>
+#include <vector>
+#include "common/common_types.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+
+namespace Vulkan {
+
+class VKDevice;
+class VKFence;
+class VKScheduler;
+class VKStagingBufferPool;
+class VKUpdateDescriptorQueue;
+
+class VKComputePass {
+public:
+    explicit VKComputePass(const VKDevice& device, VKDescriptorPool& descriptor_pool,
+                           const std::vector<vk::DescriptorSetLayoutBinding>& bindings,
+                           const std::vector<vk::DescriptorUpdateTemplateEntry>& templates,
+                           const std::vector<vk::PushConstantRange> push_constants,
+                           std::size_t code_size, const u8* code);
+    ~VKComputePass();
+
+protected:
+    vk::DescriptorSet CommitDescriptorSet(VKUpdateDescriptorQueue& update_descriptor_queue,
+                                          VKFence& fence);
+
+    UniqueDescriptorUpdateTemplate descriptor_template;
+    UniquePipelineLayout layout;
+    UniquePipeline pipeline;
+
+private:
+    UniqueDescriptorSetLayout descriptor_set_layout;
+    std::optional<DescriptorAllocator> descriptor_allocator;
+    UniqueShaderModule module;
+};
+
+class QuadArrayPass final : public VKComputePass {
+public:
+    explicit QuadArrayPass(const VKDevice& device, VKScheduler& scheduler,
+                           VKDescriptorPool& descriptor_pool,
+                           VKStagingBufferPool& staging_buffer_pool,
+                           VKUpdateDescriptorQueue& update_descriptor_queue);
+    ~QuadArrayPass();
+
+    std::pair<const vk::Buffer&, vk::DeviceSize> Assemble(u32 num_vertices, u32 first);
+
+private:
+    VKScheduler& scheduler;
+    VKStagingBufferPool& staging_buffer_pool;
+    VKUpdateDescriptorQueue& update_descriptor_queue;
+};
+
+class Uint8Pass final : public VKComputePass {
+public:
+    explicit Uint8Pass(const VKDevice& device, VKScheduler& scheduler,
+                       VKDescriptorPool& descriptor_pool, VKStagingBufferPool& staging_buffer_pool,
+                       VKUpdateDescriptorQueue& update_descriptor_queue);
+    ~Uint8Pass();
+
+    std::pair<const vk::Buffer*, u64> Assemble(u32 num_vertices, vk::Buffer src_buffer,
+                                               u64 src_offset);
+
+private:
+    VKScheduler& scheduler;
+    VKStagingBufferPool& staging_buffer_pool;
+    VKUpdateDescriptorQueue& update_descriptor_queue;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pipeline.cpp

@@ -0,0 +1,112 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <memory>
+#include <vector>
+
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
+#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
+#include "video_core/renderer_vulkan/vk_update_descriptor.h"
+
+namespace Vulkan {
+
+VKComputePipeline::VKComputePipeline(const VKDevice& device, VKScheduler& scheduler,
+                                     VKDescriptorPool& descriptor_pool,
+                                     VKUpdateDescriptorQueue& update_descriptor_queue,
+                                     const SPIRVShader& shader)
+    : device{device}, scheduler{scheduler}, entries{shader.entries},
+      descriptor_set_layout{CreateDescriptorSetLayout()},
+      descriptor_allocator{descriptor_pool, *descriptor_set_layout},
+      update_descriptor_queue{update_descriptor_queue}, layout{CreatePipelineLayout()},
+      descriptor_template{CreateDescriptorUpdateTemplate()},
+      shader_module{CreateShaderModule(shader.code)}, pipeline{CreatePipeline()} {}
+
+VKComputePipeline::~VKComputePipeline() = default;
+
+vk::DescriptorSet VKComputePipeline::CommitDescriptorSet() {
+    if (!descriptor_template) {
+        return {};
+    }
+    const auto set = descriptor_allocator.Commit(scheduler.GetFence());
+    update_descriptor_queue.Send(*descriptor_template, set);
+    return set;
+}
+
+UniqueDescriptorSetLayout VKComputePipeline::CreateDescriptorSetLayout() const {
+    std::vector<vk::DescriptorSetLayoutBinding> bindings;
+    u32 binding = 0;
+    const auto AddBindings = [&](vk::DescriptorType descriptor_type, std::size_t num_entries) {
+        // TODO(Rodrigo): Maybe make individual bindings here?
+        for (u32 bindpoint = 0; bindpoint < static_cast<u32>(num_entries); ++bindpoint) {
+            bindings.emplace_back(binding++, descriptor_type, 1, vk::ShaderStageFlagBits::eCompute,
+                                  nullptr);
+        }
+    };
+    AddBindings(vk::DescriptorType::eUniformBuffer, entries.const_buffers.size());
+    AddBindings(vk::DescriptorType::eStorageBuffer, entries.global_buffers.size());
+    AddBindings(vk::DescriptorType::eUniformTexelBuffer, entries.texel_buffers.size());
+    AddBindings(vk::DescriptorType::eCombinedImageSampler, entries.samplers.size());
+    AddBindings(vk::DescriptorType::eStorageImage, entries.images.size());
+
+    const vk::DescriptorSetLayoutCreateInfo descriptor_set_layout_ci(
+        {}, static_cast<u32>(bindings.size()), bindings.data());
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    return dev.createDescriptorSetLayoutUnique(descriptor_set_layout_ci, nullptr, dld);
+}
+
+UniquePipelineLayout VKComputePipeline::CreatePipelineLayout() const {
+    const vk::PipelineLayoutCreateInfo layout_ci({}, 1, &*descriptor_set_layout, 0, nullptr);
+    const auto dev = device.GetLogical();
+    return dev.createPipelineLayoutUnique(layout_ci, nullptr, device.GetDispatchLoader());
+}
+
+UniqueDescriptorUpdateTemplate VKComputePipeline::CreateDescriptorUpdateTemplate() const {
+    std::vector<vk::DescriptorUpdateTemplateEntry> template_entries;
+    u32 binding = 0;
+    u32 offset = 0;
+    FillDescriptorUpdateTemplateEntries(device, entries, binding, offset, template_entries);
+    if (template_entries.empty()) {
+        // If the shader doesn't use descriptor sets, skip template creation.
+        return UniqueDescriptorUpdateTemplate{};
+    }
+
+    const vk::DescriptorUpdateTemplateCreateInfo template_ci(
+        {}, static_cast<u32>(template_entries.size()), template_entries.data(),
+        vk::DescriptorUpdateTemplateType::eDescriptorSet, *descriptor_set_layout,
+        vk::PipelineBindPoint::eGraphics, *layout, DESCRIPTOR_SET);
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    return dev.createDescriptorUpdateTemplateUnique(template_ci, nullptr, dld);
+}
+
+UniqueShaderModule VKComputePipeline::CreateShaderModule(const std::vector<u32>& code) const {
+    const vk::ShaderModuleCreateInfo module_ci({}, code.size() * sizeof(u32), code.data());
+    const auto dev = device.GetLogical();
+    return dev.createShaderModuleUnique(module_ci, nullptr, device.GetDispatchLoader());
+}
+
+UniquePipeline VKComputePipeline::CreatePipeline() const {
+    vk::PipelineShaderStageCreateInfo shader_stage_ci({}, vk::ShaderStageFlagBits::eCompute,
+                                                      *shader_module, "main", nullptr);
+    vk::PipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci;
+    subgroup_size_ci.requiredSubgroupSize = GuestWarpSize;
+    if (entries.uses_warps && device.IsGuestWarpSizeSupported(vk::ShaderStageFlagBits::eCompute)) {
+        shader_stage_ci.pNext = &subgroup_size_ci;
+    }
+
+    const vk::ComputePipelineCreateInfo create_info({}, shader_stage_ci, *layout, {}, 0);
+    const auto dev = device.GetLogical();
+    return dev.createComputePipelineUnique({}, create_info, nullptr, device.GetDispatchLoader());
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pipeline.h

@@ -0,0 +1,66 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+
+#include "common/common_types.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
+
+namespace Vulkan {
+
+class VKDevice;
+class VKScheduler;
+class VKUpdateDescriptorQueue;
+
+class VKComputePipeline final {
+public:
+    explicit VKComputePipeline(const VKDevice& device, VKScheduler& scheduler,
+                               VKDescriptorPool& descriptor_pool,
+                               VKUpdateDescriptorQueue& update_descriptor_queue,
+                               const SPIRVShader& shader);
+    ~VKComputePipeline();
+
+    vk::DescriptorSet CommitDescriptorSet();
+
+    vk::Pipeline GetHandle() const {
+        return *pipeline;
+    }
+
+    vk::PipelineLayout GetLayout() const {
+        return *layout;
+    }
+
+    const ShaderEntries& GetEntries() {
+        return entries;
+    }
+
+private:
+    UniqueDescriptorSetLayout CreateDescriptorSetLayout() const;
+
+    UniquePipelineLayout CreatePipelineLayout() const;
+
+    UniqueDescriptorUpdateTemplate CreateDescriptorUpdateTemplate() const;
+
+    UniqueShaderModule CreateShaderModule(const std::vector<u32>& code) const;
+
+    UniquePipeline CreatePipeline() const;
+
+    const VKDevice& device;
+    VKScheduler& scheduler;
+    ShaderEntries entries;
+
+    UniqueDescriptorSetLayout descriptor_set_layout;
+    DescriptorAllocator descriptor_allocator;
+    VKUpdateDescriptorQueue& update_descriptor_queue;
+    UniquePipelineLayout layout;
+    UniqueDescriptorUpdateTemplate descriptor_template;
+    UniqueShaderModule shader_module;
+    UniquePipeline pipeline;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_descriptor_pool.cpp

@@ -0,0 +1,89 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <memory>
+#include <vector>
+
+#include "common/common_types.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+
+namespace Vulkan {
+
+// Prefer small grow rates to avoid saturating the descriptor pool with barely used pipelines.
+constexpr std::size_t SETS_GROW_RATE = 0x20;
+
+DescriptorAllocator::DescriptorAllocator(VKDescriptorPool& descriptor_pool,
+                                         vk::DescriptorSetLayout layout)
+    : VKFencedPool{SETS_GROW_RATE}, descriptor_pool{descriptor_pool}, layout{layout} {}
+
+DescriptorAllocator::~DescriptorAllocator() = default;
+
+vk::DescriptorSet DescriptorAllocator::Commit(VKFence& fence) {
+    return *descriptors[CommitResource(fence)];
+}
+
+void DescriptorAllocator::Allocate(std::size_t begin, std::size_t end) {
+    auto new_sets = descriptor_pool.AllocateDescriptors(layout, end - begin);
+    descriptors.insert(descriptors.end(), std::make_move_iterator(new_sets.begin()),
+                       std::make_move_iterator(new_sets.end()));
+}
+
+VKDescriptorPool::VKDescriptorPool(const VKDevice& device)
+    : device{device}, active_pool{AllocateNewPool()} {}
+
+VKDescriptorPool::~VKDescriptorPool() = default;
+
+vk::DescriptorPool VKDescriptorPool::AllocateNewPool() {
+    static constexpr u32 num_sets = 0x20000;
+    static constexpr vk::DescriptorPoolSize pool_sizes[] = {
+        {vk::DescriptorType::eUniformBuffer, num_sets * 90},
+        {vk::DescriptorType::eStorageBuffer, num_sets * 60},
+        {vk::DescriptorType::eUniformTexelBuffer, num_sets * 64},
+        {vk::DescriptorType::eCombinedImageSampler, num_sets * 64},
+        {vk::DescriptorType::eStorageImage, num_sets * 40}};
+
+    const vk::DescriptorPoolCreateInfo create_info(
+        vk::DescriptorPoolCreateFlagBits::eFreeDescriptorSet, num_sets,
+        static_cast<u32>(std::size(pool_sizes)), std::data(pool_sizes));
+    const auto dev = device.GetLogical();
+    return *pools.emplace_back(
+        dev.createDescriptorPoolUnique(create_info, nullptr, device.GetDispatchLoader()));
+}
+
+std::vector<UniqueDescriptorSet> VKDescriptorPool::AllocateDescriptors(
+    vk::DescriptorSetLayout layout, std::size_t count) {
+    std::vector layout_copies(count, layout);
+    vk::DescriptorSetAllocateInfo allocate_info(active_pool, static_cast<u32>(count),
+                                                layout_copies.data());
+
+    std::vector<vk::DescriptorSet> sets(count);
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    switch (const auto result = dev.allocateDescriptorSets(&allocate_info, sets.data(), dld)) {
+    case vk::Result::eSuccess:
+        break;
+    case vk::Result::eErrorOutOfPoolMemory:
+        active_pool = AllocateNewPool();
+        allocate_info.descriptorPool = active_pool;
+        if (dev.allocateDescriptorSets(&allocate_info, sets.data(), dld) == vk::Result::eSuccess) {
+            break;
+        }
+        [[fallthrough]];
+    default:
+        vk::throwResultException(result, "vk::Device::allocateDescriptorSetsUnique");
+    }
+
+    vk::PoolFree deleter(dev, active_pool, dld);
+    std::vector<UniqueDescriptorSet> unique_sets;
+    unique_sets.reserve(count);
+    for (const auto set : sets) {
+        unique_sets.push_back(UniqueDescriptorSet{set, deleter});
+    }
+    return unique_sets;
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_descriptor_pool.h

@@ -0,0 +1,56 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <vector>
+
+#include "common/common_types.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+
+namespace Vulkan {
+
+class VKDescriptorPool;
+
+class DescriptorAllocator final : public VKFencedPool {
+public:
+    explicit DescriptorAllocator(VKDescriptorPool& descriptor_pool, vk::DescriptorSetLayout layout);
+    ~DescriptorAllocator() override;
+
+    DescriptorAllocator(const DescriptorAllocator&) = delete;
+
+    vk::DescriptorSet Commit(VKFence& fence);
+
+protected:
+    void Allocate(std::size_t begin, std::size_t end) override;
+
+private:
+    VKDescriptorPool& descriptor_pool;
+    const vk::DescriptorSetLayout layout;
+
+    std::vector<UniqueDescriptorSet> descriptors;
+};
+
+class VKDescriptorPool final {
+    friend DescriptorAllocator;
+
+public:
+    explicit VKDescriptorPool(const VKDevice& device);
+    ~VKDescriptorPool();
+
+private:
+    vk::DescriptorPool AllocateNewPool();
+
+    std::vector<UniqueDescriptorSet> AllocateDescriptors(vk::DescriptorSetLayout layout,
+                                                         std::size_t count);
+
+    const VKDevice& device;
+
+    std::vector<UniqueDescriptorPool> pools;
+    vk::DescriptorPool active_pool;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -0,0 +1,270 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <vector>
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "common/microprofile.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
+#include "video_core/renderer_vulkan/maxwell_to_vk.h"
+#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
+#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
+#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_update_descriptor.h"
+
+namespace Vulkan {
+
+MICROPROFILE_DECLARE(Vulkan_PipelineCache);
+
+namespace {
+
+vk::StencilOpState GetStencilFaceState(const FixedPipelineState::StencilFace& face) {
+    return vk::StencilOpState(MaxwellToVK::StencilOp(face.action_stencil_fail),
+                              MaxwellToVK::StencilOp(face.action_depth_pass),
+                              MaxwellToVK::StencilOp(face.action_depth_fail),
+                              MaxwellToVK::ComparisonOp(face.test_func), 0, 0, 0);
+}
+
+bool SupportsPrimitiveRestart(vk::PrimitiveTopology topology) {
+    static constexpr std::array unsupported_topologies = {
+        vk::PrimitiveTopology::ePointList,
+        vk::PrimitiveTopology::eLineList,
+        vk::PrimitiveTopology::eTriangleList,
+        vk::PrimitiveTopology::eLineListWithAdjacency,
+        vk::PrimitiveTopology::eTriangleListWithAdjacency,
+        vk::PrimitiveTopology::ePatchList};
+    return std::find(std::begin(unsupported_topologies), std::end(unsupported_topologies),
+                     topology) == std::end(unsupported_topologies);
+}
+
+} // Anonymous namespace
+
+VKGraphicsPipeline::VKGraphicsPipeline(const VKDevice& device, VKScheduler& scheduler,
+                                       VKDescriptorPool& descriptor_pool,
+                                       VKUpdateDescriptorQueue& update_descriptor_queue,
+                                       VKRenderPassCache& renderpass_cache,
+                                       const GraphicsPipelineCacheKey& key,
+                                       const std::vector<vk::DescriptorSetLayoutBinding>& bindings,
+                                       const SPIRVProgram& program)
+    : device{device}, scheduler{scheduler}, fixed_state{key.fixed_state}, hash{key.Hash()},
+      descriptor_set_layout{CreateDescriptorSetLayout(bindings)},
+      descriptor_allocator{descriptor_pool, *descriptor_set_layout},
+      update_descriptor_queue{update_descriptor_queue}, layout{CreatePipelineLayout()},
+      descriptor_template{CreateDescriptorUpdateTemplate(program)}, modules{CreateShaderModules(
+                                                                        program)},
+      renderpass{renderpass_cache.GetRenderPass(key.renderpass_params)}, pipeline{CreatePipeline(
+                                                                             key.renderpass_params,
+                                                                             program)} {}
+
+VKGraphicsPipeline::~VKGraphicsPipeline() = default;
+
+vk::DescriptorSet VKGraphicsPipeline::CommitDescriptorSet() {
+    if (!descriptor_template) {
+        return {};
+    }
+    const auto set = descriptor_allocator.Commit(scheduler.GetFence());
+    update_descriptor_queue.Send(*descriptor_template, set);
+    return set;
+}
+
+UniqueDescriptorSetLayout VKGraphicsPipeline::CreateDescriptorSetLayout(
+    const std::vector<vk::DescriptorSetLayoutBinding>& bindings) const {
+    const vk::DescriptorSetLayoutCreateInfo descriptor_set_layout_ci(
+        {}, static_cast<u32>(bindings.size()), bindings.data());
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    return dev.createDescriptorSetLayoutUnique(descriptor_set_layout_ci, nullptr, dld);
+}
+
+UniquePipelineLayout VKGraphicsPipeline::CreatePipelineLayout() const {
+    const vk::PipelineLayoutCreateInfo pipeline_layout_ci({}, 1, &*descriptor_set_layout, 0,
+                                                          nullptr);
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    return dev.createPipelineLayoutUnique(pipeline_layout_ci, nullptr, dld);
+}
+
+UniqueDescriptorUpdateTemplate VKGraphicsPipeline::CreateDescriptorUpdateTemplate(
+    const SPIRVProgram& program) const {
+    std::vector<vk::DescriptorUpdateTemplateEntry> template_entries;
+    u32 binding = 0;
+    u32 offset = 0;
+    for (const auto& stage : program) {
+        if (stage) {
+            FillDescriptorUpdateTemplateEntries(device, stage->entries, binding, offset,
+                                                template_entries);
+        }
+    }
+    if (template_entries.empty()) {
+        // If the shader doesn't use descriptor sets, skip template creation.
+        return UniqueDescriptorUpdateTemplate{};
+    }
+
+    const vk::DescriptorUpdateTemplateCreateInfo template_ci(
+        {}, static_cast<u32>(template_entries.size()), template_entries.data(),
+        vk::DescriptorUpdateTemplateType::eDescriptorSet, *descriptor_set_layout,
+        vk::PipelineBindPoint::eGraphics, *layout, DESCRIPTOR_SET);
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    return dev.createDescriptorUpdateTemplateUnique(template_ci, nullptr, dld);
+}
+
+std::vector<UniqueShaderModule> VKGraphicsPipeline::CreateShaderModules(
+    const SPIRVProgram& program) const {
+    std::vector<UniqueShaderModule> modules;
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    for (std::size_t i = 0; i < Maxwell::MaxShaderStage; ++i) {
+        const auto& stage = program[i];
+        if (!stage) {
+            continue;
+        }
+        const vk::ShaderModuleCreateInfo module_ci({}, stage->code.size() * sizeof(u32),
+                                                   stage->code.data());
+        modules.emplace_back(dev.createShaderModuleUnique(module_ci, nullptr, dld));
+    }
+    return modules;
+}
+
+UniquePipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpass_params,
+                                                  const SPIRVProgram& program) const {
+    const auto& vi = fixed_state.vertex_input;
+    const auto& ia = fixed_state.input_assembly;
+    const auto& ds = fixed_state.depth_stencil;
+    const auto& cd = fixed_state.color_blending;
+    const auto& ts = fixed_state.tessellation;
+    const auto& rs = fixed_state.rasterizer;
+
+    std::vector<vk::VertexInputBindingDescription> vertex_bindings;
+    std::vector<vk::VertexInputBindingDivisorDescriptionEXT> vertex_binding_divisors;
+    for (std::size_t i = 0; i < vi.num_bindings; ++i) {
+        const auto& binding = vi.bindings[i];
+        const bool instanced = binding.divisor != 0;
+        const auto rate = instanced ? vk::VertexInputRate::eInstance : vk::VertexInputRate::eVertex;
+        vertex_bindings.emplace_back(binding.index, binding.stride, rate);
+        if (instanced) {
+            vertex_binding_divisors.emplace_back(binding.index, binding.divisor);
+        }
+    }
+
+    std::vector<vk::VertexInputAttributeDescription> vertex_attributes;
+    const auto& input_attributes = program[0]->entries.attributes;
+    for (std::size_t i = 0; i < vi.num_attributes; ++i) {
+        const auto& attribute = vi.attributes[i];
+        if (input_attributes.find(attribute.index) == input_attributes.end()) {
+            // Skip attributes not used by the vertex shaders.
+            continue;
+        }
+        vertex_attributes.emplace_back(attribute.index, attribute.buffer,
+                                       MaxwellToVK::VertexFormat(attribute.type, attribute.size),
+                                       attribute.offset);
+    }
+
+    vk::PipelineVertexInputStateCreateInfo vertex_input_ci(
+        {}, static_cast<u32>(vertex_bindings.size()), vertex_bindings.data(),
+        static_cast<u32>(vertex_attributes.size()), vertex_attributes.data());
+
+    const vk::PipelineVertexInputDivisorStateCreateInfoEXT vertex_input_divisor_ci(
+        static_cast<u32>(vertex_binding_divisors.size()), vertex_binding_divisors.data());
+    if (!vertex_binding_divisors.empty()) {
+        vertex_input_ci.pNext = &vertex_input_divisor_ci;
+    }
+
+    const auto primitive_topology = MaxwellToVK::PrimitiveTopology(device, ia.topology);
+    const vk::PipelineInputAssemblyStateCreateInfo input_assembly_ci(
+        {}, primitive_topology,
+        ia.primitive_restart_enable && SupportsPrimitiveRestart(primitive_topology));
+
+    const vk::PipelineTessellationStateCreateInfo tessellation_ci({}, ts.patch_control_points);
+
+    const vk::PipelineViewportStateCreateInfo viewport_ci({}, Maxwell::NumViewports, nullptr,
+                                                          Maxwell::NumViewports, nullptr);
+
+    // TODO(Rodrigo): Find out what's the default register value for front face
+    const vk::PipelineRasterizationStateCreateInfo rasterizer_ci(
+        {}, rs.depth_clamp_enable, false, vk::PolygonMode::eFill,
+        rs.cull_enable ? MaxwellToVK::CullFace(rs.cull_face) : vk::CullModeFlagBits::eNone,
+        MaxwellToVK::FrontFace(rs.front_face), rs.depth_bias_enable, 0.0f, 0.0f, 0.0f, 1.0f);
+
+    const vk::PipelineMultisampleStateCreateInfo multisampling_ci(
+        {}, vk::SampleCountFlagBits::e1, false, 0.0f, nullptr, false, false);
+
+    const vk::CompareOp depth_test_compare = ds.depth_test_enable
+                                                 ? MaxwellToVK::ComparisonOp(ds.depth_test_function)
+                                                 : vk::CompareOp::eAlways;
+
+    const vk::PipelineDepthStencilStateCreateInfo depth_stencil_ci(
+        {}, ds.depth_test_enable, ds.depth_write_enable, depth_test_compare, ds.depth_bounds_enable,
+        ds.stencil_enable, GetStencilFaceState(ds.front_stencil),
+        GetStencilFaceState(ds.back_stencil), 0.0f, 0.0f);
+
+    std::array<vk::PipelineColorBlendAttachmentState, Maxwell::NumRenderTargets> cb_attachments;
+    const std::size_t num_attachments =
+        std::min(cd.attachments_count, renderpass_params.color_attachments.size());
+    for (std::size_t i = 0; i < num_attachments; ++i) {
+        constexpr std::array component_table{
+            vk::ColorComponentFlagBits::eR, vk::ColorComponentFlagBits::eG,
+            vk::ColorComponentFlagBits::eB, vk::ColorComponentFlagBits::eA};
+        const auto& blend = cd.attachments[i];
+
+        vk::ColorComponentFlags color_components{};
+        for (std::size_t j = 0; j < component_table.size(); ++j) {
+            if (blend.components[j])
+                color_components |= component_table[j];
+        }
+
+        cb_attachments[i] = vk::PipelineColorBlendAttachmentState(
+            blend.enable, MaxwellToVK::BlendFactor(blend.src_rgb_func),
+            MaxwellToVK::BlendFactor(blend.dst_rgb_func),
+            MaxwellToVK::BlendEquation(blend.rgb_equation),
+            MaxwellToVK::BlendFactor(blend.src_a_func), MaxwellToVK::BlendFactor(blend.dst_a_func),
+            MaxwellToVK::BlendEquation(blend.a_equation), color_components);
+    }
+    const vk::PipelineColorBlendStateCreateInfo color_blending_ci({}, false, vk::LogicOp::eCopy,
+                                                                  static_cast<u32>(num_attachments),
+                                                                  cb_attachments.data(), {});
+
+    constexpr std::array dynamic_states = {
+        vk::DynamicState::eViewport,         vk::DynamicState::eScissor,
+        vk::DynamicState::eDepthBias,        vk::DynamicState::eBlendConstants,
+        vk::DynamicState::eDepthBounds,      vk::DynamicState::eStencilCompareMask,
+        vk::DynamicState::eStencilWriteMask, vk::DynamicState::eStencilReference};
+    const vk::PipelineDynamicStateCreateInfo dynamic_state_ci(
+        {}, static_cast<u32>(dynamic_states.size()), dynamic_states.data());
+
+    vk::PipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci;
+    subgroup_size_ci.requiredSubgroupSize = GuestWarpSize;
+
+    std::vector<vk::PipelineShaderStageCreateInfo> shader_stages;
+    std::size_t module_index = 0;
+    for (std::size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
+        if (!program[stage]) {
+            continue;
+        }
+        const auto stage_enum = static_cast<Tegra::Engines::ShaderType>(stage);
+        const auto vk_stage = MaxwellToVK::ShaderStage(stage_enum);
+        auto& stage_ci = shader_stages.emplace_back(vk::PipelineShaderStageCreateFlags{}, vk_stage,
+                                                    *modules[module_index++], "main", nullptr);
+        if (program[stage]->entries.uses_warps && device.IsGuestWarpSizeSupported(vk_stage)) {
+            stage_ci.pNext = &subgroup_size_ci;
+        }
+    }
+
+    const vk::GraphicsPipelineCreateInfo create_info(
+        {}, static_cast<u32>(shader_stages.size()), shader_stages.data(), &vertex_input_ci,
+        &input_assembly_ci, &tessellation_ci, &viewport_ci, &rasterizer_ci, &multisampling_ci,
+        &depth_stencil_ci, &color_blending_ci, &dynamic_state_ci, *layout, renderpass, 0, {}, 0);
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    return dev.createGraphicsPipelineUnique(nullptr, create_info, nullptr, dld);
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_graphics_pipeline.h

@@ -0,0 +1,90 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <memory>
+#include <optional>
+#include <unordered_map>
+#include <vector>
+
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
+#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
+
+namespace Vulkan {
+
+using Maxwell = Tegra::Engines::Maxwell3D::Regs;
+
+struct GraphicsPipelineCacheKey;
+
+class VKDescriptorPool;
+class VKDevice;
+class VKRenderPassCache;
+class VKScheduler;
+class VKUpdateDescriptorQueue;
+
+using SPIRVProgram = std::array<std::optional<SPIRVShader>, Maxwell::MaxShaderStage>;
+
+class VKGraphicsPipeline final {
+public:
+    explicit VKGraphicsPipeline(const VKDevice& device, VKScheduler& scheduler,
+                                VKDescriptorPool& descriptor_pool,
+                                VKUpdateDescriptorQueue& update_descriptor_queue,
+                                VKRenderPassCache& renderpass_cache,
+                                const GraphicsPipelineCacheKey& key,
+                                const std::vector<vk::DescriptorSetLayoutBinding>& bindings,
+                                const SPIRVProgram& program);
+    ~VKGraphicsPipeline();
+
+    vk::DescriptorSet CommitDescriptorSet();
+
+    vk::Pipeline GetHandle() const {
+        return *pipeline;
+    }
+
+    vk::PipelineLayout GetLayout() const {
+        return *layout;
+    }
+
+    vk::RenderPass GetRenderPass() const {
+        return renderpass;
+    }
+
+private:
+    UniqueDescriptorSetLayout CreateDescriptorSetLayout(
+        const std::vector<vk::DescriptorSetLayoutBinding>& bindings) const;
+
+    UniquePipelineLayout CreatePipelineLayout() const;
+
+    UniqueDescriptorUpdateTemplate CreateDescriptorUpdateTemplate(
+        const SPIRVProgram& program) const;
+
+    std::vector<UniqueShaderModule> CreateShaderModules(const SPIRVProgram& program) const;
+
+    UniquePipeline CreatePipeline(const RenderPassParams& renderpass_params,
+                                  const SPIRVProgram& program) const;
+
+    const VKDevice& device;
+    VKScheduler& scheduler;
+    const FixedPipelineState fixed_state;
+    const u64 hash;
+
+    UniqueDescriptorSetLayout descriptor_set_layout;
+    DescriptorAllocator descriptor_allocator;
+    VKUpdateDescriptorQueue& update_descriptor_queue;
+    UniquePipelineLayout layout;
+    UniqueDescriptorUpdateTemplate descriptor_template;
+    std::vector<UniqueShaderModule> modules;
+
+    vk::RenderPass renderpass;
+    UniquePipeline pipeline;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_memory_manager.cpp

@@ -6,6 +6,7 @@
 #include <optional>
 #include <tuple>
 #include <vector>
+
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/common_types.h"
@@ -16,34 +17,32 @@
 
 namespace Vulkan {
 
-// TODO(Rodrigo): Fine tune this number
-constexpr u64 ALLOC_CHUNK_SIZE = 64 * 1024 * 1024;
+namespace {
+
+u64 GetAllocationChunkSize(u64 required_size) {
+    static constexpr u64 sizes[] = {16ULL << 20, 32ULL << 20, 64ULL << 20, 128ULL << 20};
+    auto it = std::lower_bound(std::begin(sizes), std::end(sizes), required_size);
+    return it != std::end(sizes) ? *it : Common::AlignUp(required_size, 256ULL << 20);
+}
+
+} // Anonymous namespace
 
 class VKMemoryAllocation final {
 public:
     explicit VKMemoryAllocation(const VKDevice& device, vk::DeviceMemory memory,
-                                vk::MemoryPropertyFlags properties, u64 alloc_size, u32 type)
-        : device{device}, memory{memory}, properties{properties}, alloc_size{alloc_size},
-          shifted_type{ShiftType(type)}, is_mappable{properties &
-                                                     vk::MemoryPropertyFlagBits::eHostVisible} {
-        if (is_mappable) {
-            const auto dev = device.GetLogical();
-            const auto& dld = device.GetDispatchLoader();
-            base_address = static_cast<u8*>(dev.mapMemory(memory, 0, alloc_size, {}, dld));
-        }
-    }
+                                vk::MemoryPropertyFlags properties, u64 allocation_size, u32 type)
+        : device{device}, memory{memory}, properties{properties}, allocation_size{allocation_size},
+          shifted_type{ShiftType(type)} {}
 
     ~VKMemoryAllocation() {
         const auto dev = device.GetLogical();
         const auto& dld = device.GetDispatchLoader();
-        if (is_mappable)
-            dev.unmapMemory(memory, dld);
         dev.free(memory, nullptr, dld);
     }
 
     VKMemoryCommit Commit(vk::DeviceSize commit_size, vk::DeviceSize alignment) {
-        auto found = TryFindFreeSection(free_iterator, alloc_size, static_cast<u64>(commit_size),
-                                        static_cast<u64>(alignment));
+        auto found = TryFindFreeSection(free_iterator, allocation_size,
+                                        static_cast<u64>(commit_size), static_cast<u64>(alignment));
         if (!found) {
             found = TryFindFreeSection(0, free_iterator, static_cast<u64>(commit_size),
                                        static_cast<u64>(alignment));
@@ -52,8 +51,7 @@ public:
                 return nullptr;
             }
         }
-        u8* address = is_mappable ? base_address + *found : nullptr;
-        auto commit = std::make_unique<VKMemoryCommitImpl>(this, memory, address, *found,
+        auto commit = std::make_unique<VKMemoryCommitImpl>(device, this, memory, *found,
                                                            *found + commit_size);
         commits.push_back(commit.get());
 
@@ -65,12 +63,10 @@ public:
 
     void Free(const VKMemoryCommitImpl* commit) {
         ASSERT(commit);
-        const auto it =
-            std::find_if(commits.begin(), commits.end(),
-                         [&](const auto& stored_commit) { return stored_commit == commit; });
+
+        const auto it = std::find(std::begin(commits), std::end(commits), commit);
         if (it == commits.end()) {
-            LOG_CRITICAL(Render_Vulkan, "Freeing unallocated commit!");
-            UNREACHABLE();
+            UNREACHABLE_MSG("Freeing unallocated commit!");
             return;
         }
         commits.erase(it);
@@ -88,11 +84,11 @@ private:
     }
 
     /// A memory allocator, it may return a free region between "start" and "end" with the solicited
-    /// requeriments.
+    /// requirements.
     std::optional<u64> TryFindFreeSection(u64 start, u64 end, u64 size, u64 alignment) const {
-        u64 iterator = start;
-        while (iterator + size < end) {
-            const u64 try_left = Common::AlignUp(iterator, alignment);
+        u64 iterator = Common::AlignUp(start, alignment);
+        while (iterator + size <= end) {
+            const u64 try_left = iterator;
             const u64 try_right = try_left + size;
 
             bool overlap = false;
@@ -100,7 +96,7 @@ private:
                 const auto [commit_left, commit_right] = commit->interval;
                 if (try_left < commit_right && commit_left < try_right) {
                     // There's an overlap, continue the search where the overlapping commit ends.
-                    iterator = commit_right;
+                    iterator = Common::AlignUp(commit_right, alignment);
                     overlap = true;
                     break;
                 }
@@ -110,6 +106,7 @@ private:
                 return try_left;
             }
         }
+
         // No free regions where found, return an empty optional.
         return std::nullopt;
     }
@@ -117,12 +114,8 @@ private:
     const VKDevice& device;                   ///< Vulkan device.
     const vk::DeviceMemory memory;            ///< Vulkan memory allocation handler.
     const vk::MemoryPropertyFlags properties; ///< Vulkan properties.
-    const u64 alloc_size;                     ///< Size of this allocation.
+    const u64 allocation_size;                ///< Size of this allocation.
     const u32 shifted_type;                   ///< Stored Vulkan type of this allocation, shifted.
-    const bool is_mappable;                   ///< Whether the allocation is mappable.
-
-    /// Base address of the mapped pointer.
-    u8* base_address{};
 
     /// Hints where the next free region is likely going to be.
     u64 free_iterator{};
@@ -132,13 +125,15 @@ private:
 };
 
 VKMemoryManager::VKMemoryManager(const VKDevice& device)
-    : device{device}, props{device.GetPhysical().getMemoryProperties(device.GetDispatchLoader())},
-      is_memory_unified{GetMemoryUnified(props)} {}
+    : device{device}, properties{device.GetPhysical().getMemoryProperties(
+                          device.GetDispatchLoader())},
+      is_memory_unified{GetMemoryUnified(properties)} {}
 
 VKMemoryManager::~VKMemoryManager() = default;
 
-VKMemoryCommit VKMemoryManager::Commit(const vk::MemoryRequirements& reqs, bool host_visible) {
-    ASSERT(reqs.size < ALLOC_CHUNK_SIZE);
+VKMemoryCommit VKMemoryManager::Commit(const vk::MemoryRequirements& requirements,
+                                       bool host_visible) {
+    const u64 chunk_size = GetAllocationChunkSize(requirements.size);
 
     // When a host visible commit is asked, search for host visible and coherent, otherwise search
     // for a fast device local type.
@@ -147,32 +142,21 @@ VKMemoryCommit VKMemoryManager::Commit(const vk::MemoryRequirements& reqs, bool
             ? vk::MemoryPropertyFlagBits::eHostVisible | vk::MemoryPropertyFlagBits::eHostCoherent
             : vk::MemoryPropertyFlagBits::eDeviceLocal;
 
-    const auto TryCommit = [&]() -> VKMemoryCommit {
-        for (auto& alloc : allocs) {
-            if (!alloc->IsCompatible(wanted_properties, reqs.memoryTypeBits))
-                continue;
-
-            if (auto commit = alloc->Commit(reqs.size, reqs.alignment); commit) {
-                return commit;
-            }
-        }
-        return {};
-    };
-
-    if (auto commit = TryCommit(); commit) {
+    if (auto commit = TryAllocCommit(requirements, wanted_properties)) {
         return commit;
     }
 
     // Commit has failed, allocate more memory.
-    if (!AllocMemory(wanted_properties, reqs.memoryTypeBits, ALLOC_CHUNK_SIZE)) {
-        // TODO(Rodrigo): Try to use host memory.
-        LOG_CRITICAL(Render_Vulkan, "Ran out of memory!");
-        UNREACHABLE();
+    if (!AllocMemory(wanted_properties, requirements.memoryTypeBits, chunk_size)) {
+        // TODO(Rodrigo): Handle these situations in some way like flushing to guest memory.
+        // Allocation has failed, panic.
+        UNREACHABLE_MSG("Ran out of VRAM!");
+        return {};
     }
 
     // Commit again, this time it won't fail since there's a fresh allocation above. If it does,
     // there's a bug.
-    auto commit = TryCommit();
+    auto commit = TryAllocCommit(requirements, wanted_properties);
     ASSERT(commit);
     return commit;
 }
@@ -180,8 +164,7 @@ VKMemoryCommit VKMemoryManager::Commit(const vk::MemoryRequirements& reqs, bool
 VKMemoryCommit VKMemoryManager::Commit(vk::Buffer buffer, bool host_visible) {
     const auto dev = device.GetLogical();
     const auto& dld = device.GetDispatchLoader();
-    const auto requeriments = dev.getBufferMemoryRequirements(buffer, dld);
-    auto commit = Commit(requeriments, host_visible);
+    auto commit = Commit(dev.getBufferMemoryRequirements(buffer, dld), host_visible);
     dev.bindBufferMemory(buffer, commit->GetMemory(), commit->GetOffset(), dld);
     return commit;
 }
@@ -189,25 +172,23 @@ VKMemoryCommit VKMemoryManager::Commit(vk::Buffer buffer, bool host_visible) {
 VKMemoryCommit VKMemoryManager::Commit(vk::Image image, bool host_visible) {
     const auto dev = device.GetLogical();
     const auto& dld = device.GetDispatchLoader();
-    const auto requeriments = dev.getImageMemoryRequirements(image, dld);
-    auto commit = Commit(requeriments, host_visible);
+    auto commit = Commit(dev.getImageMemoryRequirements(image, dld), host_visible);
     dev.bindImageMemory(image, commit->GetMemory(), commit->GetOffset(), dld);
     return commit;
 }
 
 bool VKMemoryManager::AllocMemory(vk::MemoryPropertyFlags wanted_properties, u32 type_mask,
                                   u64 size) {
-    const u32 type = [&]() {
-        for (u32 type_index = 0; type_index < props.memoryTypeCount; ++type_index) {
-            const auto flags = props.memoryTypes[type_index].propertyFlags;
+    const u32 type = [&] {
+        for (u32 type_index = 0; type_index < properties.memoryTypeCount; ++type_index) {
+            const auto flags = properties.memoryTypes[type_index].propertyFlags;
             if ((type_mask & (1U << type_index)) && (flags & wanted_properties)) {
                 // The type matches in type and in the wanted properties.
                 return type_index;
             }
         }
-        LOG_CRITICAL(Render_Vulkan, "Couldn't find a compatible memory type!");
-        UNREACHABLE();
-        return 0u;
+        UNREACHABLE_MSG("Couldn't find a compatible memory type!");
+        return 0U;
     }();
 
     const auto dev = device.GetLogical();
@@ -216,19 +197,33 @@ bool VKMemoryManager::AllocMemory(vk::MemoryPropertyFlags wanted_properties, u32
     // Try to allocate found type.
     const vk::MemoryAllocateInfo memory_ai(size, type);
     vk::DeviceMemory memory;
-    if (const vk::Result res = dev.allocateMemory(&memory_ai, nullptr, &memory, dld);
+    if (const auto res = dev.allocateMemory(&memory_ai, nullptr, &memory, dld);
         res != vk::Result::eSuccess) {
         LOG_CRITICAL(Render_Vulkan, "Device allocation failed with code {}!", vk::to_string(res));
         return false;
     }
-    allocs.push_back(
+    allocations.push_back(
         std::make_unique<VKMemoryAllocation>(device, memory, wanted_properties, size, type));
     return true;
 }
 
-/*static*/ bool VKMemoryManager::GetMemoryUnified(const vk::PhysicalDeviceMemoryProperties& props) {
-    for (u32 heap_index = 0; heap_index < props.memoryHeapCount; ++heap_index) {
-        if (!(props.memoryHeaps[heap_index].flags & vk::MemoryHeapFlagBits::eDeviceLocal)) {
+VKMemoryCommit VKMemoryManager::TryAllocCommit(const vk::MemoryRequirements& requirements,
+                                               vk::MemoryPropertyFlags wanted_properties) {
+    for (auto& allocation : allocations) {
+        if (!allocation->IsCompatible(wanted_properties, requirements.memoryTypeBits)) {
+            continue;
+        }
+        if (auto commit = allocation->Commit(requirements.size, requirements.alignment)) {
+            return commit;
+        }
+    }
+    return {};
+}
+
+/*static*/ bool VKMemoryManager::GetMemoryUnified(
+    const vk::PhysicalDeviceMemoryProperties& properties) {
+    for (u32 heap_index = 0; heap_index < properties.memoryHeapCount; ++heap_index) {
+        if (!(properties.memoryHeaps[heap_index].flags & vk::MemoryHeapFlagBits::eDeviceLocal)) {
             // Memory is considered unified when heaps are device local only.
             return false;
         }
@@ -236,17 +231,28 @@ bool VKMemoryManager::AllocMemory(vk::MemoryPropertyFlags wanted_properties, u32
     return true;
 }
 
-VKMemoryCommitImpl::VKMemoryCommitImpl(VKMemoryAllocation* allocation, vk::DeviceMemory memory,
-                                       u8* data, u64 begin, u64 end)
-    : interval(std::make_pair(begin, end)), memory{memory}, allocation{allocation}, data{data} {}
+VKMemoryCommitImpl::VKMemoryCommitImpl(const VKDevice& device, VKMemoryAllocation* allocation,
+                                       vk::DeviceMemory memory, u64 begin, u64 end)
+    : device{device}, interval{begin, end}, memory{memory}, allocation{allocation} {}
 
 VKMemoryCommitImpl::~VKMemoryCommitImpl() {
     allocation->Free(this);
 }
 
-u8* VKMemoryCommitImpl::GetData() const {
-    ASSERT_MSG(data != nullptr, "Trying to access an unmapped commit.");
-    return data;
+MemoryMap VKMemoryCommitImpl::Map(u64 size, u64 offset_) const {
+    const auto dev = device.GetLogical();
+    const auto address = reinterpret_cast<u8*>(
+        dev.mapMemory(memory, interval.first + offset_, size, {}, device.GetDispatchLoader()));
+    return MemoryMap{this, address};
+}
+
+void VKMemoryCommitImpl::Unmap() const {
+    const auto dev = device.GetLogical();
+    dev.unmapMemory(memory, device.GetDispatchLoader());
+}
+
+MemoryMap VKMemoryCommitImpl::Map() const {
+    return Map(interval.second - interval.first);
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_memory_manager.h

@@ -12,6 +12,7 @@
 
 namespace Vulkan {
 
+class MemoryMap;
 class VKDevice;
 class VKMemoryAllocation;
 class VKMemoryCommitImpl;
@@ -21,13 +22,14 @@ using VKMemoryCommit = std::unique_ptr<VKMemoryCommitImpl>;
 class VKMemoryManager final {
 public:
     explicit VKMemoryManager(const VKDevice& device);
+    VKMemoryManager(const VKMemoryManager&) = delete;
     ~VKMemoryManager();
 
     /**
      * Commits a memory with the specified requeriments.
-     * @param reqs Requeriments returned from a Vulkan call.
+     * @param requirements Requirements returned from a Vulkan call.
      * @param host_visible Signals the allocator that it *must* use host visible and coherent
-     * memory. When passing false, it will try to allocate device local memory.
+     *                     memory. When passing false, it will try to allocate device local memory.
      * @returns A memory commit.
      */
     VKMemoryCommit Commit(const vk::MemoryRequirements& reqs, bool host_visible);
@@ -47,25 +49,35 @@ private:
     /// Allocates a chunk of memory.
     bool AllocMemory(vk::MemoryPropertyFlags wanted_properties, u32 type_mask, u64 size);
 
-    /// Returns true if the device uses an unified memory model.
-    static bool GetMemoryUnified(const vk::PhysicalDeviceMemoryProperties& props);
+    /// Tries to allocate a memory commit.
+    VKMemoryCommit TryAllocCommit(const vk::MemoryRequirements& requirements,
+                                  vk::MemoryPropertyFlags wanted_properties);
 
-    const VKDevice& device;                                  ///< Device handler.
-    const vk::PhysicalDeviceMemoryProperties props;          ///< Physical device properties.
-    const bool is_memory_unified;                            ///< True if memory model is unified.
-    std::vector<std::unique_ptr<VKMemoryAllocation>> allocs; ///< Current allocations.
+    /// Returns true if the device uses an unified memory model.
+    static bool GetMemoryUnified(const vk::PhysicalDeviceMemoryProperties& properties);
+
+    const VKDevice& device;                              ///< Device handler.
+    const vk::PhysicalDeviceMemoryProperties properties; ///< Physical device properties.
+    const bool is_memory_unified;                        ///< True if memory model is unified.
+    std::vector<std::unique_ptr<VKMemoryAllocation>> allocations; ///< Current allocations.
 };
 
 class VKMemoryCommitImpl final {
     friend VKMemoryAllocation;
+    friend MemoryMap;
 
 public:
-    explicit VKMemoryCommitImpl(VKMemoryAllocation* allocation, vk::DeviceMemory memory, u8* data,
-                                u64 begin, u64 end);
+    explicit VKMemoryCommitImpl(const VKDevice& device, VKMemoryAllocation* allocation,
+                                vk::DeviceMemory memory, u64 begin, u64 end);
     ~VKMemoryCommitImpl();
 
-    /// Returns the writeable memory map. The commit has to be mappable.
-    u8* GetData() const;
+    /// Maps a memory region and returns a pointer to it.
+    /// It's illegal to have more than one memory map at the same time.
+    MemoryMap Map(u64 size, u64 offset = 0) const;
+
+    /// Maps the whole commit and returns a pointer to it.
+    /// It's illegal to have more than one memory map at the same time.
+    MemoryMap Map() const;
 
     /// Returns the Vulkan memory handler.
     vk::DeviceMemory GetMemory() const {
@@ -78,10 +90,46 @@ public:
     }
 
 private:
+    /// Unmaps memory.
+    void Unmap() const;
+
+    const VKDevice& device;           ///< Vulkan device.
     std::pair<u64, u64> interval{};   ///< Interval where the commit exists.
     vk::DeviceMemory memory;          ///< Vulkan device memory handler.
     VKMemoryAllocation* allocation{}; ///< Pointer to the large memory allocation.
-    u8* data{}; ///< Pointer to the host mapped memory, it has the commit offset included.
+};
+
+/// Holds ownership of a memory map.
+class MemoryMap final {
+public:
+    explicit MemoryMap(const VKMemoryCommitImpl* commit, u8* address)
+        : commit{commit}, address{address} {}
+
+    ~MemoryMap() {
+        if (commit) {
+            commit->Unmap();
+        }
+    }
+
+    /// Prematurely releases the memory map.
+    void Release() {
+        commit->Unmap();
+        commit = nullptr;
+    }
+
+    /// Returns the address of the memory map.
+    u8* GetAddress() const {
+        return address;
+    }
+
+    /// Returns the address of the memory map;
+    operator u8*() const {
+        return address;
+    }
+
+private:
+    const VKMemoryCommitImpl* commit{}; ///< Mapped memory commit.
+    u8* address{};                      ///< Address to the mapped memory.
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -0,0 +1,392 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <cstddef>
+#include <memory>
+#include <vector>
+
+#include "common/microprofile.h"
+#include "core/core.h"
+#include "core/memory.h"
+#include "video_core/engines/kepler_compute.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/memory_manager.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
+#include "video_core/renderer_vulkan/maxwell_to_vk.h"
+#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
+#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
+#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
+#include "video_core/renderer_vulkan/vk_rasterizer.h"
+#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_update_descriptor.h"
+#include "video_core/shader/compiler_settings.h"
+
+namespace Vulkan {
+
+MICROPROFILE_DECLARE(Vulkan_PipelineCache);
+
+using Tegra::Engines::ShaderType;
+
+namespace {
+
+constexpr VideoCommon::Shader::CompilerSettings compiler_settings{
+    VideoCommon::Shader::CompileDepth::FullDecompile};
+
+/// Gets the address for the specified shader stage program
+GPUVAddr GetShaderAddress(Core::System& system, Maxwell::ShaderProgram program) {
+    const auto& gpu{system.GPU().Maxwell3D()};
+    const auto& shader_config{gpu.regs.shader_config[static_cast<std::size_t>(program)]};
+    return gpu.regs.code_address.CodeAddress() + shader_config.offset;
+}
+
+/// Gets if the current instruction offset is a scheduler instruction
+constexpr bool IsSchedInstruction(std::size_t offset, std::size_t main_offset) {
+    // Sched instructions appear once every 4 instructions.
+    constexpr std::size_t SchedPeriod = 4;
+    const std::size_t absolute_offset = offset - main_offset;
+    return (absolute_offset % SchedPeriod) == 0;
+}
+
+/// Calculates the size of a program stream
+std::size_t CalculateProgramSize(const ProgramCode& program, bool is_compute) {
+    const std::size_t start_offset = is_compute ? 0 : 10;
+    // This is the encoded version of BRA that jumps to itself. All Nvidia
+    // shaders end with one.
+    constexpr u64 self_jumping_branch = 0xE2400FFFFF07000FULL;
+    constexpr u64 mask = 0xFFFFFFFFFF7FFFFFULL;
+    std::size_t offset = start_offset;
+    while (offset < program.size()) {
+        const u64 instruction = program[offset];
+        if (!IsSchedInstruction(offset, start_offset)) {
+            if ((instruction & mask) == self_jumping_branch) {
+                // End on Maxwell's "nop" instruction
+                break;
+            }
+            if (instruction == 0) {
+                break;
+            }
+        }
+        ++offset;
+    }
+    // The last instruction is included in the program size
+    return std::min(offset + 1, program.size());
+}
+
+/// Gets the shader program code from memory for the specified address
+ProgramCode GetShaderCode(Tegra::MemoryManager& memory_manager, const GPUVAddr gpu_addr,
+                          const u8* host_ptr, bool is_compute) {
+    ProgramCode program_code(VideoCommon::Shader::MAX_PROGRAM_LENGTH);
+    ASSERT_OR_EXECUTE(host_ptr != nullptr, {
+        std::fill(program_code.begin(), program_code.end(), 0);
+        return program_code;
+    });
+    memory_manager.ReadBlockUnsafe(gpu_addr, program_code.data(),
+                                   program_code.size() * sizeof(u64));
+    program_code.resize(CalculateProgramSize(program_code, is_compute));
+    return program_code;
+}
+
+constexpr std::size_t GetStageFromProgram(std::size_t program) {
+    return program == 0 ? 0 : program - 1;
+}
+
+constexpr ShaderType GetStageFromProgram(Maxwell::ShaderProgram program) {
+    return static_cast<ShaderType>(GetStageFromProgram(static_cast<std::size_t>(program)));
+}
+
+ShaderType GetShaderType(Maxwell::ShaderProgram program) {
+    switch (program) {
+    case Maxwell::ShaderProgram::VertexB:
+        return ShaderType::Vertex;
+    case Maxwell::ShaderProgram::TesselationControl:
+        return ShaderType::TesselationControl;
+    case Maxwell::ShaderProgram::TesselationEval:
+        return ShaderType::TesselationEval;
+    case Maxwell::ShaderProgram::Geometry:
+        return ShaderType::Geometry;
+    case Maxwell::ShaderProgram::Fragment:
+        return ShaderType::Fragment;
+    default:
+        UNIMPLEMENTED_MSG("program={}", static_cast<u32>(program));
+        return ShaderType::Vertex;
+    }
+}
+
+u32 FillDescriptorLayout(const ShaderEntries& entries,
+                         std::vector<vk::DescriptorSetLayoutBinding>& bindings,
+                         Maxwell::ShaderProgram program_type, u32 base_binding) {
+    const ShaderType stage = GetStageFromProgram(program_type);
+    const vk::ShaderStageFlags stage_flags = MaxwellToVK::ShaderStage(stage);
+
+    u32 binding = base_binding;
+    const auto AddBindings = [&](vk::DescriptorType descriptor_type, std::size_t num_entries) {
+        for (std::size_t i = 0; i < num_entries; ++i) {
+            bindings.emplace_back(binding++, descriptor_type, 1, stage_flags, nullptr);
+        }
+    };
+    AddBindings(vk::DescriptorType::eUniformBuffer, entries.const_buffers.size());
+    AddBindings(vk::DescriptorType::eStorageBuffer, entries.global_buffers.size());
+    AddBindings(vk::DescriptorType::eUniformTexelBuffer, entries.texel_buffers.size());
+    AddBindings(vk::DescriptorType::eCombinedImageSampler, entries.samplers.size());
+    AddBindings(vk::DescriptorType::eStorageImage, entries.images.size());
+    return binding;
+}
+
+} // Anonymous namespace
+
+CachedShader::CachedShader(Core::System& system, Tegra::Engines::ShaderType stage,
+                           GPUVAddr gpu_addr, VAddr cpu_addr, u8* host_ptr,
+                           ProgramCode program_code, u32 main_offset)
+    : RasterizerCacheObject{host_ptr}, gpu_addr{gpu_addr}, cpu_addr{cpu_addr},
+      program_code{std::move(program_code)}, locker{stage, GetEngine(system, stage)},
+      shader_ir{this->program_code, main_offset, compiler_settings, locker},
+      entries{GenerateShaderEntries(shader_ir)} {}
+
+CachedShader::~CachedShader() = default;
+
+Tegra::Engines::ConstBufferEngineInterface& CachedShader::GetEngine(
+    Core::System& system, Tegra::Engines::ShaderType stage) {
+    if (stage == Tegra::Engines::ShaderType::Compute) {
+        return system.GPU().KeplerCompute();
+    } else {
+        return system.GPU().Maxwell3D();
+    }
+}
+
+VKPipelineCache::VKPipelineCache(Core::System& system, RasterizerVulkan& rasterizer,
+                                 const VKDevice& device, VKScheduler& scheduler,
+                                 VKDescriptorPool& descriptor_pool,
+                                 VKUpdateDescriptorQueue& update_descriptor_queue)
+    : RasterizerCache{rasterizer}, system{system}, device{device}, scheduler{scheduler},
+      descriptor_pool{descriptor_pool}, update_descriptor_queue{update_descriptor_queue},
+      renderpass_cache(device) {}
+
+VKPipelineCache::~VKPipelineCache() = default;
+
+std::array<Shader, Maxwell::MaxShaderProgram> VKPipelineCache::GetShaders() {
+    const auto& gpu = system.GPU().Maxwell3D();
+    auto& dirty = system.GPU().Maxwell3D().dirty.shaders;
+    if (!dirty) {
+        return last_shaders;
+    }
+    dirty = false;
+
+    std::array<Shader, Maxwell::MaxShaderProgram> shaders;
+    for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
+        const auto& shader_config = gpu.regs.shader_config[index];
+        const auto program{static_cast<Maxwell::ShaderProgram>(index)};
+
+        // Skip stages that are not enabled
+        if (!gpu.regs.IsShaderConfigEnabled(index)) {
+            continue;
+        }
+
+        auto& memory_manager{system.GPU().MemoryManager()};
+        const GPUVAddr program_addr{GetShaderAddress(system, program)};
+        const auto host_ptr{memory_manager.GetPointer(program_addr)};
+        auto shader = TryGet(host_ptr);
+        if (!shader) {
+            // No shader found - create a new one
+            constexpr u32 stage_offset = 10;
+            const auto stage = static_cast<Tegra::Engines::ShaderType>(index == 0 ? 0 : index - 1);
+            auto code = GetShaderCode(memory_manager, program_addr, host_ptr, false);
+
+            const std::optional cpu_addr = memory_manager.GpuToCpuAddress(program_addr);
+            ASSERT(cpu_addr);
+
+            shader = std::make_shared<CachedShader>(system, stage, program_addr, *cpu_addr,
+                                                    host_ptr, std::move(code), stage_offset);
+            Register(shader);
+        }
+        shaders[index] = std::move(shader);
+    }
+    return last_shaders = shaders;
+}
+
+VKGraphicsPipeline& VKPipelineCache::GetGraphicsPipeline(const GraphicsPipelineCacheKey& key) {
+    MICROPROFILE_SCOPE(Vulkan_PipelineCache);
+
+    if (last_graphics_pipeline && last_graphics_key == key) {
+        return *last_graphics_pipeline;
+    }
+    last_graphics_key = key;
+
+    const auto [pair, is_cache_miss] = graphics_cache.try_emplace(key);
+    auto& entry = pair->second;
+    if (is_cache_miss) {
+        LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
+        const auto [program, bindings] = DecompileShaders(key);
+        entry = std::make_unique<VKGraphicsPipeline>(device, scheduler, descriptor_pool,
+                                                     update_descriptor_queue, renderpass_cache, key,
+                                                     bindings, program);
+    }
+    return *(last_graphics_pipeline = entry.get());
+}
+
+VKComputePipeline& VKPipelineCache::GetComputePipeline(const ComputePipelineCacheKey& key) {
+    MICROPROFILE_SCOPE(Vulkan_PipelineCache);
+
+    const auto [pair, is_cache_miss] = compute_cache.try_emplace(key);
+    auto& entry = pair->second;
+    if (!is_cache_miss) {
+        return *entry;
+    }
+    LOG_INFO(Render_Vulkan, "Compile 0x{:016X}", key.Hash());
+
+    auto& memory_manager = system.GPU().MemoryManager();
+    const auto program_addr = key.shader;
+    const auto host_ptr = memory_manager.GetPointer(program_addr);
+
+    auto shader = TryGet(host_ptr);
+    if (!shader) {
+        // No shader found - create a new one
+        const auto cpu_addr = memory_manager.GpuToCpuAddress(program_addr);
+        ASSERT(cpu_addr);
+
+        auto code = GetShaderCode(memory_manager, program_addr, host_ptr, true);
+        constexpr u32 kernel_main_offset = 0;
+        shader = std::make_shared<CachedShader>(system, Tegra::Engines::ShaderType::Compute,
+                                                program_addr, *cpu_addr, host_ptr, std::move(code),
+                                                kernel_main_offset);
+        Register(shader);
+    }
+
+    Specialization specialization;
+    specialization.workgroup_size = key.workgroup_size;
+    specialization.shared_memory_size = key.shared_memory_size;
+
+    const SPIRVShader spirv_shader{
+        Decompile(device, shader->GetIR(), ShaderType::Compute, specialization),
+        shader->GetEntries()};
+    entry = std::make_unique<VKComputePipeline>(device, scheduler, descriptor_pool,
+                                                update_descriptor_queue, spirv_shader);
+    return *entry;
+}
+
+void VKPipelineCache::Unregister(const Shader& shader) {
+    bool finished = false;
+    const auto Finish = [&] {
+        // TODO(Rodrigo): Instead of finishing here, wait for the fences that use this pipeline and
+        // flush.
+        if (finished) {
+            return;
+        }
+        finished = true;
+        scheduler.Finish();
+    };
+
+    const GPUVAddr invalidated_addr = shader->GetGpuAddr();
+    for (auto it = graphics_cache.begin(); it != graphics_cache.end();) {
+        auto& entry = it->first;
+        if (std::find(entry.shaders.begin(), entry.shaders.end(), invalidated_addr) ==
+            entry.shaders.end()) {
+            ++it;
+            continue;
+        }
+        Finish();
+        it = graphics_cache.erase(it);
+    }
+    for (auto it = compute_cache.begin(); it != compute_cache.end();) {
+        auto& entry = it->first;
+        if (entry.shader != invalidated_addr) {
+            ++it;
+            continue;
+        }
+        Finish();
+        it = compute_cache.erase(it);
+    }
+
+    RasterizerCache::Unregister(shader);
+}
+
+std::pair<SPIRVProgram, std::vector<vk::DescriptorSetLayoutBinding>>
+VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
+    const auto& fixed_state = key.fixed_state;
+    auto& memory_manager = system.GPU().MemoryManager();
+    const auto& gpu = system.GPU().Maxwell3D();
+
+    Specialization specialization;
+    specialization.primitive_topology = fixed_state.input_assembly.topology;
+    if (specialization.primitive_topology == Maxwell::PrimitiveTopology::Points) {
+        ASSERT(fixed_state.input_assembly.point_size != 0.0f);
+        specialization.point_size = fixed_state.input_assembly.point_size;
+    }
+    for (std::size_t i = 0; i < Maxwell::NumVertexAttributes; ++i) {
+        specialization.attribute_types[i] = fixed_state.vertex_input.attributes[i].type;
+    }
+    specialization.ndc_minus_one_to_one = fixed_state.rasterizer.ndc_minus_one_to_one;
+    specialization.tessellation.primitive = fixed_state.tessellation.primitive;
+    specialization.tessellation.spacing = fixed_state.tessellation.spacing;
+    specialization.tessellation.clockwise = fixed_state.tessellation.clockwise;
+
+    SPIRVProgram program;
+    std::vector<vk::DescriptorSetLayoutBinding> bindings;
+
+    for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
+        const auto program_enum = static_cast<Maxwell::ShaderProgram>(index);
+
+        // Skip stages that are not enabled
+        if (!gpu.regs.IsShaderConfigEnabled(index)) {
+            continue;
+        }
+
+        const GPUVAddr gpu_addr = GetShaderAddress(system, program_enum);
+        const auto host_ptr = memory_manager.GetPointer(gpu_addr);
+        const auto shader = TryGet(host_ptr);
+        ASSERT(shader);
+
+        const std::size_t stage = index == 0 ? 0 : index - 1; // Stage indices are 0 - 5
+        const auto program_type = GetShaderType(program_enum);
+        const auto& entries = shader->GetEntries();
+        program[stage] = {Decompile(device, shader->GetIR(), program_type, specialization),
+                          entries};
+
+        if (program_enum == Maxwell::ShaderProgram::VertexA) {
+            // VertexB was combined with VertexA, so we skip the VertexB iteration
+            ++index;
+        }
+
+        const u32 old_binding = specialization.base_binding;
+        specialization.base_binding =
+            FillDescriptorLayout(entries, bindings, program_enum, specialization.base_binding);
+        ASSERT(old_binding + entries.NumBindings() == specialization.base_binding);
+    }
+    return {std::move(program), std::move(bindings)};
+}
+
+void FillDescriptorUpdateTemplateEntries(
+    const VKDevice& device, const ShaderEntries& entries, u32& binding, u32& offset,
+    std::vector<vk::DescriptorUpdateTemplateEntry>& template_entries) {
+    static constexpr auto entry_size = static_cast<u32>(sizeof(DescriptorUpdateEntry));
+    const auto AddEntry = [&](vk::DescriptorType descriptor_type, std::size_t count_) {
+        const u32 count = static_cast<u32>(count_);
+        if (descriptor_type == vk::DescriptorType::eUniformTexelBuffer &&
+            device.GetDriverID() == vk::DriverIdKHR::eNvidiaProprietary) {
+            // Nvidia has a bug where updating multiple uniform texels at once causes the driver to
+            // crash.
+            for (u32 i = 0; i < count; ++i) {
+                template_entries.emplace_back(binding + i, 0, 1, descriptor_type,
+                                              offset + i * entry_size, entry_size);
+            }
+        } else if (count != 0) {
+            template_entries.emplace_back(binding, 0, count, descriptor_type, offset, entry_size);
+        }
+        offset += count * entry_size;
+        binding += count;
+    };
+
+    AddEntry(vk::DescriptorType::eUniformBuffer, entries.const_buffers.size());
+    AddEntry(vk::DescriptorType::eStorageBuffer, entries.global_buffers.size());
+    AddEntry(vk::DescriptorType::eUniformTexelBuffer, entries.texel_buffers.size());
+    AddEntry(vk::DescriptorType::eCombinedImageSampler, entries.samplers.size());
+    AddEntry(vk::DescriptorType::eStorageImage, entries.images.size());
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_pipeline_cache.h

@@ -0,0 +1,200 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <cstddef>
+#include <memory>
+#include <tuple>
+#include <type_traits>
+#include <unordered_map>
+#include <utility>
+#include <vector>
+
+#include <boost/functional/hash.hpp>
+
+#include "common/common_types.h"
+#include "video_core/engines/const_buffer_engine_interface.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/rasterizer_cache.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
+#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
+#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
+#include "video_core/shader/const_buffer_locker.h"
+#include "video_core/shader/shader_ir.h"
+#include "video_core/surface.h"
+
+namespace Core {
+class System;
+}
+
+namespace Vulkan {
+
+class RasterizerVulkan;
+class VKComputePipeline;
+class VKDescriptorPool;
+class VKDevice;
+class VKFence;
+class VKScheduler;
+class VKUpdateDescriptorQueue;
+
+class CachedShader;
+using Shader = std::shared_ptr<CachedShader>;
+using Maxwell = Tegra::Engines::Maxwell3D::Regs;
+
+using ProgramCode = std::vector<u64>;
+
+struct GraphicsPipelineCacheKey {
+    FixedPipelineState fixed_state;
+    std::array<GPUVAddr, Maxwell::MaxShaderProgram> shaders;
+    RenderPassParams renderpass_params;
+
+    std::size_t Hash() const noexcept {
+        std::size_t hash = fixed_state.Hash();
+        for (const auto& shader : shaders) {
+            boost::hash_combine(hash, shader);
+        }
+        boost::hash_combine(hash, renderpass_params.Hash());
+        return hash;
+    }
+
+    bool operator==(const GraphicsPipelineCacheKey& rhs) const noexcept {
+        return std::tie(fixed_state, shaders, renderpass_params) ==
+               std::tie(rhs.fixed_state, rhs.shaders, rhs.renderpass_params);
+    }
+};
+
+struct ComputePipelineCacheKey {
+    GPUVAddr shader{};
+    u32 shared_memory_size{};
+    std::array<u32, 3> workgroup_size{};
+
+    std::size_t Hash() const noexcept {
+        return static_cast<std::size_t>(shader) ^
+               ((static_cast<std::size_t>(shared_memory_size) >> 7) << 40) ^
+               static_cast<std::size_t>(workgroup_size[0]) ^
+               (static_cast<std::size_t>(workgroup_size[1]) << 16) ^
+               (static_cast<std::size_t>(workgroup_size[2]) << 24);
+    }
+
+    bool operator==(const ComputePipelineCacheKey& rhs) const noexcept {
+        return std::tie(shader, shared_memory_size, workgroup_size) ==
+               std::tie(rhs.shader, rhs.shared_memory_size, rhs.workgroup_size);
+    }
+};
+
+} // namespace Vulkan
+
+namespace std {
+
+template <>
+struct hash<Vulkan::GraphicsPipelineCacheKey> {
+    std::size_t operator()(const Vulkan::GraphicsPipelineCacheKey& k) const noexcept {
+        return k.Hash();
+    }
+};
+
+template <>
+struct hash<Vulkan::ComputePipelineCacheKey> {
+    std::size_t operator()(const Vulkan::ComputePipelineCacheKey& k) const noexcept {
+        return k.Hash();
+    }
+};
+
+} // namespace std
+
+namespace Vulkan {
+
+class CachedShader final : public RasterizerCacheObject {
+public:
+    explicit CachedShader(Core::System& system, Tegra::Engines::ShaderType stage, GPUVAddr gpu_addr,
+                          VAddr cpu_addr, u8* host_ptr, ProgramCode program_code, u32 main_offset);
+    ~CachedShader();
+
+    GPUVAddr GetGpuAddr() const {
+        return gpu_addr;
+    }
+
+    VAddr GetCpuAddr() const override {
+        return cpu_addr;
+    }
+
+    std::size_t GetSizeInBytes() const override {
+        return program_code.size() * sizeof(u64);
+    }
+
+    VideoCommon::Shader::ShaderIR& GetIR() {
+        return shader_ir;
+    }
+
+    const VideoCommon::Shader::ShaderIR& GetIR() const {
+        return shader_ir;
+    }
+
+    const ShaderEntries& GetEntries() const {
+        return entries;
+    }
+
+private:
+    static Tegra::Engines::ConstBufferEngineInterface& GetEngine(Core::System& system,
+                                                                 Tegra::Engines::ShaderType stage);
+
+    GPUVAddr gpu_addr{};
+    VAddr cpu_addr{};
+    ProgramCode program_code;
+    VideoCommon::Shader::ConstBufferLocker locker;
+    VideoCommon::Shader::ShaderIR shader_ir;
+    ShaderEntries entries;
+};
+
+class VKPipelineCache final : public RasterizerCache<Shader> {
+public:
+    explicit VKPipelineCache(Core::System& system, RasterizerVulkan& rasterizer,
+                             const VKDevice& device, VKScheduler& scheduler,
+                             VKDescriptorPool& descriptor_pool,
+                             VKUpdateDescriptorQueue& update_descriptor_queue);
+    ~VKPipelineCache();
+
+    std::array<Shader, Maxwell::MaxShaderProgram> GetShaders();
+
+    VKGraphicsPipeline& GetGraphicsPipeline(const GraphicsPipelineCacheKey& key);
+
+    VKComputePipeline& GetComputePipeline(const ComputePipelineCacheKey& key);
+
+protected:
+    void Unregister(const Shader& shader) override;
+
+    void FlushObjectInner(const Shader& object) override {}
+
+private:
+    std::pair<SPIRVProgram, std::vector<vk::DescriptorSetLayoutBinding>> DecompileShaders(
+        const GraphicsPipelineCacheKey& key);
+
+    Core::System& system;
+    const VKDevice& device;
+    VKScheduler& scheduler;
+    VKDescriptorPool& descriptor_pool;
+    VKUpdateDescriptorQueue& update_descriptor_queue;
+
+    VKRenderPassCache renderpass_cache;
+
+    std::array<Shader, Maxwell::MaxShaderProgram> last_shaders;
+
+    GraphicsPipelineCacheKey last_graphics_key;
+    VKGraphicsPipeline* last_graphics_pipeline = nullptr;
+
+    std::unordered_map<GraphicsPipelineCacheKey, std::unique_ptr<VKGraphicsPipeline>>
+        graphics_cache;
+    std::unordered_map<ComputePipelineCacheKey, std::unique_ptr<VKComputePipeline>> compute_cache;
+};
+
+void FillDescriptorUpdateTemplateEntries(
+    const VKDevice& device, const ShaderEntries& entries, u32& binding, u32& offset,
+    std::vector<vk::DescriptorUpdateTemplateEntry>& template_entries);
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -0,0 +1,263 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <bitset>
+#include <memory>
+#include <utility>
+#include <vector>
+
+#include <boost/container/static_vector.hpp>
+#include <boost/functional/hash.hpp>
+
+#include "common/common_types.h"
+#include "video_core/memory_manager.h"
+#include "video_core/rasterizer_accelerated.h"
+#include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
+#include "video_core/renderer_vulkan/vk_buffer_cache.h"
+#include "video_core/renderer_vulkan/vk_compute_pass.h"
+#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
+#include "video_core/renderer_vulkan/vk_memory_manager.h"
+#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
+#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+#include "video_core/renderer_vulkan/vk_sampler_cache.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
+#include "video_core/renderer_vulkan/vk_texture_cache.h"
+#include "video_core/renderer_vulkan/vk_update_descriptor.h"
+
+namespace Core {
+class System;
+}
+
+namespace Core::Frontend {
+class EmuWindow;
+}
+
+namespace Tegra::Engines {
+class Maxwell3D;
+}
+
+namespace Vulkan {
+
+struct VKScreenInfo;
+
+using ImageViewsPack =
+    boost::container::static_vector<vk::ImageView, Maxwell::NumRenderTargets + 1>;
+
+struct FramebufferCacheKey {
+    vk::RenderPass renderpass{};
+    u32 width = 0;
+    u32 height = 0;
+    ImageViewsPack views;
+
+    std::size_t Hash() const noexcept {
+        std::size_t hash = 0;
+        boost::hash_combine(hash, static_cast<VkRenderPass>(renderpass));
+        for (const auto& view : views) {
+            boost::hash_combine(hash, static_cast<VkImageView>(view));
+        }
+        boost::hash_combine(hash, width);
+        boost::hash_combine(hash, height);
+        return hash;
+    }
+
+    bool operator==(const FramebufferCacheKey& rhs) const noexcept {
+        return std::tie(renderpass, views, width, height) ==
+               std::tie(rhs.renderpass, rhs.views, rhs.width, rhs.height);
+    }
+};
+
+} // namespace Vulkan
+
+namespace std {
+
+template <>
+struct hash<Vulkan::FramebufferCacheKey> {
+    std::size_t operator()(const Vulkan::FramebufferCacheKey& k) const noexcept {
+        return k.Hash();
+    }
+};
+
+} // namespace std
+
+namespace Vulkan {
+
+class BufferBindings;
+
+struct ImageView {
+    View view;
+    vk::ImageLayout* layout = nullptr;
+};
+
+class RasterizerVulkan : public VideoCore::RasterizerAccelerated {
+public:
+    explicit RasterizerVulkan(Core::System& system, Core::Frontend::EmuWindow& render_window,
+                              VKScreenInfo& screen_info, const VKDevice& device,
+                              VKResourceManager& resource_manager, VKMemoryManager& memory_manager,
+                              VKScheduler& scheduler);
+    ~RasterizerVulkan() override;
+
+    bool DrawBatch(bool is_indexed) override;
+    bool DrawMultiBatch(bool is_indexed) override;
+    void Clear() override;
+    void DispatchCompute(GPUVAddr code_addr) override;
+    void FlushAll() override;
+    void FlushRegion(CacheAddr addr, u64 size) override;
+    void InvalidateRegion(CacheAddr addr, u64 size) override;
+    void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
+    void FlushCommands() override;
+    void TickFrame() override;
+    bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
+                               const Tegra::Engines::Fermi2D::Regs::Surface& dst,
+                               const Tegra::Engines::Fermi2D::Config& copy_config) override;
+    bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
+                           u32 pixel_stride) override;
+
+    /// Maximum supported size that a constbuffer can have in bytes.
+    static constexpr std::size_t MaxConstbufferSize = 0x10000;
+    static_assert(MaxConstbufferSize % (4 * sizeof(float)) == 0,
+                  "The maximum size of a constbuffer must be a multiple of the size of GLvec4");
+
+private:
+    struct DrawParameters {
+        void Draw(vk::CommandBuffer cmdbuf, const vk::DispatchLoaderDynamic& dld) const;
+
+        u32 base_instance = 0;
+        u32 num_instances = 0;
+        u32 base_vertex = 0;
+        u32 num_vertices = 0;
+        bool is_indexed = 0;
+    };
+
+    using Texceptions = std::bitset<Maxwell::NumRenderTargets + 1>;
+
+    static constexpr std::size_t ZETA_TEXCEPTION_INDEX = 8;
+
+    void Draw(bool is_indexed, bool is_instanced);
+
+    void FlushWork();
+
+    Texceptions UpdateAttachments();
+
+    std::tuple<vk::Framebuffer, vk::Extent2D> ConfigureFramebuffers(vk::RenderPass renderpass);
+
+    /// Setups geometry buffers and state.
+    DrawParameters SetupGeometry(FixedPipelineState& fixed_state, BufferBindings& buffer_bindings,
+                                 bool is_indexed, bool is_instanced);
+
+    /// Setup descriptors in the graphics pipeline.
+    void SetupShaderDescriptors(const std::array<Shader, Maxwell::MaxShaderProgram>& shaders);
+
+    void SetupImageTransitions(Texceptions texceptions,
+                               const std::array<View, Maxwell::NumRenderTargets>& color_attachments,
+                               const View& zeta_attachment);
+
+    void UpdateDynamicStates();
+
+    bool WalkAttachmentOverlaps(const CachedSurfaceView& attachment);
+
+    void SetupVertexArrays(FixedPipelineState::VertexInput& vertex_input,
+                           BufferBindings& buffer_bindings);
+
+    void SetupIndexBuffer(BufferBindings& buffer_bindings, DrawParameters& params, bool is_indexed);
+
+    /// Setup constant buffers in the graphics pipeline.
+    void SetupGraphicsConstBuffers(const ShaderEntries& entries, std::size_t stage);
+
+    /// Setup global buffers in the graphics pipeline.
+    void SetupGraphicsGlobalBuffers(const ShaderEntries& entries, std::size_t stage);
+
+    /// Setup texel buffers in the graphics pipeline.
+    void SetupGraphicsTexelBuffers(const ShaderEntries& entries, std::size_t stage);
+
+    /// Setup textures in the graphics pipeline.
+    void SetupGraphicsTextures(const ShaderEntries& entries, std::size_t stage);
+
+    /// Setup images in the graphics pipeline.
+    void SetupGraphicsImages(const ShaderEntries& entries, std::size_t stage);
+
+    /// Setup constant buffers in the compute pipeline.
+    void SetupComputeConstBuffers(const ShaderEntries& entries);
+
+    /// Setup global buffers in the compute pipeline.
+    void SetupComputeGlobalBuffers(const ShaderEntries& entries);
+
+    /// Setup texel buffers in the compute pipeline.
+    void SetupComputeTexelBuffers(const ShaderEntries& entries);
+
+    /// Setup textures in the compute pipeline.
+    void SetupComputeTextures(const ShaderEntries& entries);
+
+    /// Setup images in the compute pipeline.
+    void SetupComputeImages(const ShaderEntries& entries);
+
+    void SetupConstBuffer(const ConstBufferEntry& entry,
+                          const Tegra::Engines::ConstBufferInfo& buffer);
+
+    void SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAddr address);
+
+    void SetupTexelBuffer(const Tegra::Texture::TICEntry& image, const TexelBufferEntry& entry);
+
+    void SetupTexture(const Tegra::Texture::FullTextureInfo& texture, const SamplerEntry& entry);
+
+    void SetupImage(const Tegra::Texture::TICEntry& tic, const ImageEntry& entry);
+
+    void UpdateViewportsState(Tegra::Engines::Maxwell3D& gpu);
+    void UpdateScissorsState(Tegra::Engines::Maxwell3D& gpu);
+    void UpdateDepthBias(Tegra::Engines::Maxwell3D& gpu);
+    void UpdateBlendConstants(Tegra::Engines::Maxwell3D& gpu);
+    void UpdateDepthBounds(Tegra::Engines::Maxwell3D& gpu);
+    void UpdateStencilFaces(Tegra::Engines::Maxwell3D& gpu);
+
+    std::size_t CalculateGraphicsStreamBufferSize(bool is_indexed) const;
+
+    std::size_t CalculateComputeStreamBufferSize() const;
+
+    std::size_t CalculateVertexArraysSize() const;
+
+    std::size_t CalculateIndexBufferSize() const;
+
+    std::size_t CalculateConstBufferSize(const ConstBufferEntry& entry,
+                                         const Tegra::Engines::ConstBufferInfo& buffer) const;
+
+    RenderPassParams GetRenderPassParams(Texceptions texceptions) const;
+
+    Core::System& system;
+    Core::Frontend::EmuWindow& render_window;
+    VKScreenInfo& screen_info;
+    const VKDevice& device;
+    VKResourceManager& resource_manager;
+    VKMemoryManager& memory_manager;
+    VKScheduler& scheduler;
+
+    VKStagingBufferPool staging_pool;
+    VKDescriptorPool descriptor_pool;
+    VKUpdateDescriptorQueue update_descriptor_queue;
+    QuadArrayPass quad_array_pass;
+    Uint8Pass uint8_pass;
+
+    VKTextureCache texture_cache;
+    VKPipelineCache pipeline_cache;
+    VKBufferCache buffer_cache;
+    VKSamplerCache sampler_cache;
+
+    std::array<View, Maxwell::NumRenderTargets> color_attachments;
+    View zeta_attachment;
+
+    std::vector<ImageView> sampled_views;
+    std::vector<ImageView> image_views;
+
+    u32 draw_counter = 0;
+
+    // TODO(Rodrigo): Invalidate on image destruction
+    std::unordered_map<FramebufferCacheKey, UniqueFramebuffer> framebuffer_cache;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_renderpass_cache.cpp

@@ -0,0 +1,100 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <memory>
+#include <vector>
+
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/maxwell_to_vk.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
+
+namespace Vulkan {
+
+VKRenderPassCache::VKRenderPassCache(const VKDevice& device) : device{device} {}
+
+VKRenderPassCache::~VKRenderPassCache() = default;
+
+vk::RenderPass VKRenderPassCache::GetRenderPass(const RenderPassParams& params) {
+    const auto [pair, is_cache_miss] = cache.try_emplace(params);
+    auto& entry = pair->second;
+    if (is_cache_miss) {
+        entry = CreateRenderPass(params);
+    }
+    return *entry;
+}
+
+UniqueRenderPass VKRenderPassCache::CreateRenderPass(const RenderPassParams& params) const {
+    std::vector<vk::AttachmentDescription> descriptors;
+    std::vector<vk::AttachmentReference> color_references;
+
+    for (std::size_t rt = 0; rt < params.color_attachments.size(); ++rt) {
+        const auto attachment = params.color_attachments[rt];
+        const auto format =
+            MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, attachment.pixel_format);
+        ASSERT_MSG(format.attachable, "Trying to attach a non-attachable format with format={}",
+                   static_cast<u32>(attachment.pixel_format));
+
+        // TODO(Rodrigo): Add eMayAlias when it's needed.
+        const auto color_layout = attachment.is_texception
+                                      ? vk::ImageLayout::eGeneral
+                                      : vk::ImageLayout::eColorAttachmentOptimal;
+        descriptors.emplace_back(vk::AttachmentDescriptionFlagBits::eMayAlias, format.format,
+                                 vk::SampleCountFlagBits::e1, vk::AttachmentLoadOp::eLoad,
+                                 vk::AttachmentStoreOp::eStore, vk::AttachmentLoadOp::eDontCare,
+                                 vk::AttachmentStoreOp::eDontCare, color_layout, color_layout);
+        color_references.emplace_back(static_cast<u32>(rt), color_layout);
+    }
+
+    vk::AttachmentReference zeta_attachment_ref;
+    if (params.has_zeta) {
+        const auto format =
+            MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, params.zeta_pixel_format);
+        ASSERT_MSG(format.attachable, "Trying to attach a non-attachable format with format={}",
+                   static_cast<u32>(params.zeta_pixel_format));
+
+        const auto zeta_layout = params.zeta_texception
+                                     ? vk::ImageLayout::eGeneral
+                                     : vk::ImageLayout::eDepthStencilAttachmentOptimal;
+        descriptors.emplace_back(vk::AttachmentDescriptionFlags{}, format.format,
+                                 vk::SampleCountFlagBits::e1, vk::AttachmentLoadOp::eLoad,
+                                 vk::AttachmentStoreOp::eStore, vk::AttachmentLoadOp::eLoad,
+                                 vk::AttachmentStoreOp::eStore, zeta_layout, zeta_layout);
+        zeta_attachment_ref =
+            vk::AttachmentReference(static_cast<u32>(params.color_attachments.size()), zeta_layout);
+    }
+
+    const vk::SubpassDescription subpass_description(
+        {}, vk::PipelineBindPoint::eGraphics, 0, nullptr, static_cast<u32>(color_references.size()),
+        color_references.data(), nullptr, params.has_zeta ? &zeta_attachment_ref : nullptr, 0,
+        nullptr);
+
+    vk::AccessFlags access;
+    vk::PipelineStageFlags stage;
+    if (!color_references.empty()) {
+        access |=
+            vk::AccessFlagBits::eColorAttachmentRead | vk::AccessFlagBits::eColorAttachmentWrite;
+        stage |= vk::PipelineStageFlagBits::eColorAttachmentOutput;
+    }
+
+    if (params.has_zeta) {
+        access |= vk::AccessFlagBits::eDepthStencilAttachmentRead |
+                  vk::AccessFlagBits::eDepthStencilAttachmentWrite;
+        stage |= vk::PipelineStageFlagBits::eLateFragmentTests;
+    }
+
+    const vk::SubpassDependency subpass_dependency(VK_SUBPASS_EXTERNAL, 0, stage, stage, {}, access,
+                                                   {});
+
+    const vk::RenderPassCreateInfo create_info({}, static_cast<u32>(descriptors.size()),
+                                               descriptors.data(), 1, &subpass_description, 1,
+                                               &subpass_dependency);
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    return dev.createRenderPassUnique(create_info, nullptr, dld);
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_renderpass_cache.h

@@ -0,0 +1,97 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <tuple>
+#include <unordered_map>
+
+#include <boost/container/static_vector.hpp>
+#include <boost/functional/hash.hpp>
+
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/surface.h"
+
+namespace Vulkan {
+
+class VKDevice;
+
+// TODO(Rodrigo): Optimize this structure for faster hashing
+
+struct RenderPassParams {
+    struct ColorAttachment {
+        u32 index = 0;
+        VideoCore::Surface::PixelFormat pixel_format = VideoCore::Surface::PixelFormat::Invalid;
+        bool is_texception = false;
+
+        std::size_t Hash() const noexcept {
+            return static_cast<std::size_t>(pixel_format) |
+                   static_cast<std::size_t>(is_texception) << 6 |
+                   static_cast<std::size_t>(index) << 7;
+        }
+
+        bool operator==(const ColorAttachment& rhs) const noexcept {
+            return std::tie(index, pixel_format, is_texception) ==
+                   std::tie(rhs.index, rhs.pixel_format, rhs.is_texception);
+        }
+    };
+
+    boost::container::static_vector<ColorAttachment,
+                                    Tegra::Engines::Maxwell3D::Regs::NumRenderTargets>
+        color_attachments{};
+    // TODO(Rodrigo): Unify has_zeta into zeta_pixel_format and zeta_component_type.
+    VideoCore::Surface::PixelFormat zeta_pixel_format = VideoCore::Surface::PixelFormat::Invalid;
+    bool has_zeta = false;
+    bool zeta_texception = false;
+
+    std::size_t Hash() const noexcept {
+        std::size_t hash = 0;
+        for (const auto& rt : color_attachments) {
+            boost::hash_combine(hash, rt.Hash());
+        }
+        boost::hash_combine(hash, zeta_pixel_format);
+        boost::hash_combine(hash, has_zeta);
+        boost::hash_combine(hash, zeta_texception);
+        return hash;
+    }
+
+    bool operator==(const RenderPassParams& rhs) const {
+        return std::tie(color_attachments, zeta_pixel_format, has_zeta, zeta_texception) ==
+               std::tie(rhs.color_attachments, rhs.zeta_pixel_format, rhs.has_zeta,
+                        rhs.zeta_texception);
+    }
+};
+
+} // namespace Vulkan
+
+namespace std {
+
+template <>
+struct hash<Vulkan::RenderPassParams> {
+    std::size_t operator()(const Vulkan::RenderPassParams& k) const noexcept {
+        return k.Hash();
+    }
+};
+
+} // namespace std
+
+namespace Vulkan {
+
+class VKRenderPassCache final {
+public:
+    explicit VKRenderPassCache(const VKDevice& device);
+    ~VKRenderPassCache();
+
+    vk::RenderPass GetRenderPass(const RenderPassParams& params);
+
+private:
+    UniqueRenderPass CreateRenderPass(const RenderPassParams& params) const;
+
+    const VKDevice& device;
+    std::unordered_map<RenderPassParams, UniqueRenderPass> cache;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_sampler_cache.cpp

@@ -46,9 +46,9 @@ UniqueSampler VKSamplerCache::CreateSampler(const Tegra::Texture::TSCEntry& tsc)
         {}, MaxwellToVK::Sampler::Filter(tsc.mag_filter),
         MaxwellToVK::Sampler::Filter(tsc.min_filter),
         MaxwellToVK::Sampler::MipmapMode(tsc.mipmap_filter),
-        MaxwellToVK::Sampler::WrapMode(tsc.wrap_u, tsc.mag_filter),
-        MaxwellToVK::Sampler::WrapMode(tsc.wrap_v, tsc.mag_filter),
-        MaxwellToVK::Sampler::WrapMode(tsc.wrap_p, tsc.mag_filter), tsc.GetLodBias(),
+        MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_u, tsc.mag_filter),
+        MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_v, tsc.mag_filter),
+        MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_p, tsc.mag_filter), tsc.GetLodBias(),
         has_anisotropy, max_anisotropy, tsc.depth_compare_enabled,
         MaxwellToVK::Sampler::DepthCompareFunction(tsc.depth_compare_func), tsc.GetMinLod(),
         tsc.GetMaxLod(), vk_border_color.value_or(vk::BorderColor::eFloatTransparentBlack),

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -542,11 +542,10 @@ private:
             return;
         }
 
-        for (u32 rt = 0; rt < static_cast<u32>(frag_colors.size()); ++rt) {
-            if (!specialization.enabled_rendertargets[rt]) {
+        for (u32 rt = 0; rt < static_cast<u32>(std::size(frag_colors)); ++rt) {
+            if (!IsRenderTargetEnabled(rt)) {
                 continue;
             }
-
             const Id id = AddGlobalVariable(OpVariable(t_out_float4, spv::StorageClass::Output));
             Name(id, fmt::format("frag_color{}", rt));
             Decorate(id, spv::Decoration::Location, rt);
@@ -852,6 +851,15 @@ private:
         return binding;
     }
 
+    bool IsRenderTargetEnabled(u32 rt) const {
+        for (u32 component = 0; component < 4; ++component) {
+            if (header.ps.IsColorComponentOutputEnabled(rt, component)) {
+                return true;
+            }
+        }
+        return false;
+    }
+
     bool IsInputAttributeArray() const {
         return stage == ShaderType::TesselationControl || stage == ShaderType::TesselationEval ||
                stage == ShaderType::Geometry;
@@ -954,6 +962,10 @@ private:
 
     Expression Visit(const Node& node) {
         if (const auto operation = std::get_if<OperationNode>(&*node)) {
+            if (const auto amend_index = operation->GetAmendIndex()) {
+                [[maybe_unused]] const Type type = Visit(ir.GetAmendNode(*amend_index)).type;
+                ASSERT(type == Type::Void);
+            }
             const auto operation_index = static_cast<std::size_t>(operation->GetCode());
             const auto decompiler = operation_decompilers[operation_index];
             if (decompiler == nullptr) {
@@ -1142,6 +1154,10 @@ private:
         }
 
         if (const auto conditional = std::get_if<ConditionalNode>(&*node)) {
+            if (const auto amend_index = conditional->GetAmendIndex()) {
+                [[maybe_unused]] const Type type = Visit(ir.GetAmendNode(*amend_index)).type;
+                ASSERT(type == Type::Void);
+            }
             // It's invalid to call conditional on nested nodes, use an operation instead
             const Id true_label = OpLabel();
             const Id skip_label = OpLabel();
@@ -1788,6 +1804,19 @@ private:
         return {};
     }
 
+    Expression UAtomicAdd(Operation operation) {
+        const auto& smem = std::get<SmemNode>(*operation[0]);
+        Id address = AsUint(Visit(smem.GetAddress()));
+        address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
+        const Id pointer = OpAccessChain(t_smem_uint, shared_memory, address);
+
+        const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
+        const Id semantics = Constant(t_uint, 0U);
+
+        const Id value = AsUint(Visit(operation[1]));
+        return {OpAtomicIAdd(t_uint, pointer, scope, semantics, value), Type::Uint};
+    }
+
     Expression Branch(Operation operation) {
         const auto& target = std::get<ImmediateNode>(*operation[0]);
         OpStore(jmp_to, Constant(t_uint, target.GetValue()));
@@ -1868,19 +1897,14 @@ private:
             // rendertargets/components are skipped in the register assignment.
             u32 current_reg = 0;
             for (u32 rt = 0; rt < Maxwell::NumRenderTargets; ++rt) {
-                if (!specialization.enabled_rendertargets[rt]) {
-                    // Skip rendertargets that are not enabled
-                    continue;
-                }
                 // TODO(Subv): Figure out how dual-source blending is configured in the Switch.
                 for (u32 component = 0; component < 4; ++component) {
-                    const Id pointer = AccessElement(t_out_float, frag_colors.at(rt), component);
-                    if (header.ps.IsColorComponentOutputEnabled(rt, component)) {
-                        OpStore(pointer, SafeGetRegister(current_reg));
-                        ++current_reg;
-                    } else {
-                        OpStore(pointer, component == 3 ? v_float_one : v_float_zero);
+                    if (!header.ps.IsColorComponentOutputEnabled(rt, component)) {
+                        continue;
                     }
+                    const Id pointer = AccessElement(t_out_float, frag_colors[rt], component);
+                    OpStore(pointer, SafeGetRegister(current_reg));
+                    ++current_reg;
                 }
             }
             if (header.ps.omap.depth) {
@@ -2365,6 +2389,8 @@ private:
         &SPIRVDecompiler::AtomicImageXor,
         &SPIRVDecompiler::AtomicImageExchange,
 
+        &SPIRVDecompiler::UAtomicAdd,
+
         &SPIRVDecompiler::Branch,
         &SPIRVDecompiler::BranchIndirect,
         &SPIRVDecompiler::PushFlowStack,

src/video_core/renderer_vulkan/vk_shader_decompiler.h

@@ -102,9 +102,6 @@ struct Specialization final {
         Maxwell::TessellationSpacing spacing{};
         bool clockwise{};
     } tessellation;
-
-    // Fragment specific
-    std::bitset<8> enabled_rendertargets;
 };
 // Old gcc versions don't consider this trivially copyable.
 // static_assert(std::is_trivially_copyable_v<Specialization>);

src/video_core/renderer_vulkan/vk_shader_util.cpp

@@ -0,0 +1,34 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <cstring>
+#include <memory>
+#include <vector>
+#include "common/alignment.h"
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_shader_util.h"
+
+namespace Vulkan {
+
+UniqueShaderModule BuildShader(const VKDevice& device, std::size_t code_size, const u8* code_data) {
+    // Avoid undefined behavior by copying to a staging allocation
+    ASSERT(code_size % sizeof(u32) == 0);
+    const auto data = std::make_unique<u32[]>(code_size / sizeof(u32));
+    std::memcpy(data.get(), code_data, code_size);
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    const vk::ShaderModuleCreateInfo shader_ci({}, code_size, data.get());
+    vk::ShaderModule shader_module;
+    if (dev.createShaderModule(&shader_ci, nullptr, &shader_module, dld) != vk::Result::eSuccess) {
+        UNREACHABLE_MSG("Shader module failed to build!");
+    }
+
+    return UniqueShaderModule(shader_module, vk::ObjectDestroy(dev, nullptr, dld));
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_shader_util.h

@@ -0,0 +1,17 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <vector>
+#include "common/common_types.h"
+#include "video_core/renderer_vulkan/declarations.h"
+
+namespace Vulkan {
+
+class VKDevice;
+
+UniqueShaderModule BuildShader(const VKDevice& device, std::size_t code_size, const u8* code_data);
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_staging_buffer_pool.h

@@ -13,6 +13,7 @@
 
 #include "video_core/renderer_vulkan/declarations.h"
 #include "video_core/renderer_vulkan/vk_memory_manager.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
 
 namespace Vulkan {
 

src/video_core/renderer_vulkan/vk_stream_buffer.cpp

@@ -3,86 +3,144 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
-#include <memory>
 #include <optional>
+#include <tuple>
 #include <vector>
 
+#include "common/alignment.h"
 #include "common/assert.h"
 #include "video_core/renderer_vulkan/declarations.h"
 #include "video_core/renderer_vulkan/vk_device.h"
-#include "video_core/renderer_vulkan/vk_memory_manager.h"
 #include "video_core/renderer_vulkan/vk_resource_manager.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_stream_buffer.h"
 
 namespace Vulkan {
 
+namespace {
+
 constexpr u64 WATCHES_INITIAL_RESERVE = 0x4000;
 constexpr u64 WATCHES_RESERVE_CHUNK = 0x1000;
 
-VKStreamBuffer::VKStreamBuffer(const VKDevice& device, VKMemoryManager& memory_manager,
-                               VKScheduler& scheduler, u64 size, vk::BufferUsageFlags usage,
-                               vk::AccessFlags access, vk::PipelineStageFlags pipeline_stage)
-    : device{device}, scheduler{scheduler}, buffer_size{size}, access{access}, pipeline_stage{
-                                                                                   pipeline_stage} {
-    CreateBuffers(memory_manager, usage);
-    ReserveWatches(WATCHES_INITIAL_RESERVE);
+constexpr u64 STREAM_BUFFER_SIZE = 256 * 1024 * 1024;
+
+std::optional<u32> FindMemoryType(const VKDevice& device, u32 filter,
+                                  vk::MemoryPropertyFlags wanted) {
+    const auto properties = device.GetPhysical().getMemoryProperties(device.GetDispatchLoader());
+    for (u32 i = 0; i < properties.memoryTypeCount; i++) {
+        if (!(filter & (1 << i))) {
+            continue;
+        }
+        if ((properties.memoryTypes[i].propertyFlags & wanted) == wanted) {
+            return i;
+        }
+    }
+    return {};
+}
+
+} // Anonymous namespace
+
+VKStreamBuffer::VKStreamBuffer(const VKDevice& device, VKScheduler& scheduler,
+                               vk::BufferUsageFlags usage)
+    : device{device}, scheduler{scheduler} {
+    CreateBuffers(usage);
+    ReserveWatches(current_watches, WATCHES_INITIAL_RESERVE);
+    ReserveWatches(previous_watches, WATCHES_INITIAL_RESERVE);
 }
 
 VKStreamBuffer::~VKStreamBuffer() = default;
 
-std::tuple<u8*, u64, bool> VKStreamBuffer::Reserve(u64 size) {
-    ASSERT(size <= buffer_size);
+std::tuple<u8*, u64, bool> VKStreamBuffer::Map(u64 size, u64 alignment) {
+    ASSERT(size <= STREAM_BUFFER_SIZE);
     mapped_size = size;
 
-    if (offset + size > buffer_size) {
-        // The buffer would overflow, save the amount of used buffers, signal an invalidation and
-        // reset the state.
-        invalidation_mark = used_watches;
-        used_watches = 0;
+    if (alignment > 0) {
+        offset = Common::AlignUp(offset, alignment);
+    }
+
+    WaitPendingOperations(offset);
+
+    bool invalidated = false;
+    if (offset + size > STREAM_BUFFER_SIZE) {
+        // The buffer would overflow, save the amount of used watches and reset the state.
+        invalidation_mark = current_watch_cursor;
+        current_watch_cursor = 0;
         offset = 0;
-    }
 
-    return {mapped_pointer + offset, offset, invalidation_mark.has_value()};
-}
+        // Swap watches and reset waiting cursors.
+        std::swap(previous_watches, current_watches);
+        wait_cursor = 0;
+        wait_bound = 0;
 
-void VKStreamBuffer::Send(u64 size) {
-    ASSERT_MSG(size <= mapped_size, "Reserved size is too small");
-
-    if (invalidation_mark) {
-        // TODO(Rodrigo): Find a better way to invalidate than waiting for all watches to finish.
+        // Ensure that we don't wait for uncommitted fences.
         scheduler.Flush();
-        std::for_each(watches.begin(), watches.begin() + *invalidation_mark,
-                      [&](auto& resource) { resource->Wait(); });
-        invalidation_mark = std::nullopt;
+
+        invalidated = true;
     }
 
-    if (used_watches + 1 >= watches.size()) {
-        // Ensure that there are enough watches.
-        ReserveWatches(WATCHES_RESERVE_CHUNK);
-    }
-    // Add a watch for this allocation.
-    watches[used_watches++]->Watch(scheduler.GetFence());
-
-    offset += size;
-}
-
-void VKStreamBuffer::CreateBuffers(VKMemoryManager& memory_manager, vk::BufferUsageFlags usage) {
-    const vk::BufferCreateInfo buffer_ci({}, buffer_size, usage, vk::SharingMode::eExclusive, 0,
-                                         nullptr);
-
     const auto dev = device.GetLogical();
     const auto& dld = device.GetDispatchLoader();
-    buffer = dev.createBufferUnique(buffer_ci, nullptr, dld);
-    commit = memory_manager.Commit(*buffer, true);
-    mapped_pointer = commit->GetData();
+    const auto pointer = reinterpret_cast<u8*>(dev.mapMemory(*memory, offset, size, {}, dld));
+    return {pointer, offset, invalidated};
 }
 
-void VKStreamBuffer::ReserveWatches(std::size_t grow_size) {
-    const std::size_t previous_size = watches.size();
-    watches.resize(previous_size + grow_size);
-    std::generate(watches.begin() + previous_size, watches.end(),
-                  []() { return std::make_unique<VKFenceWatch>(); });
+void VKStreamBuffer::Unmap(u64 size) {
+    ASSERT_MSG(size <= mapped_size, "Reserved size is too small");
+
+    const auto dev = device.GetLogical();
+    dev.unmapMemory(*memory, device.GetDispatchLoader());
+
+    offset += size;
+
+    if (current_watch_cursor + 1 >= current_watches.size()) {
+        // Ensure that there are enough watches.
+        ReserveWatches(current_watches, WATCHES_RESERVE_CHUNK);
+    }
+    auto& watch = current_watches[current_watch_cursor++];
+    watch.upper_bound = offset;
+    watch.fence.Watch(scheduler.GetFence());
+}
+
+void VKStreamBuffer::CreateBuffers(vk::BufferUsageFlags usage) {
+    const vk::BufferCreateInfo buffer_ci({}, STREAM_BUFFER_SIZE, usage, vk::SharingMode::eExclusive,
+                                         0, nullptr);
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    buffer = dev.createBufferUnique(buffer_ci, nullptr, dld);
+
+    const auto requirements = dev.getBufferMemoryRequirements(*buffer, dld);
+    // Prefer device local host visible allocations (this should hit AMD's pinned memory).
+    auto type = FindMemoryType(device, requirements.memoryTypeBits,
+                               vk::MemoryPropertyFlagBits::eHostVisible |
+                                   vk::MemoryPropertyFlagBits::eHostCoherent |
+                                   vk::MemoryPropertyFlagBits::eDeviceLocal);
+    if (!type) {
+        // Otherwise search for a host visible allocation.
+        type = FindMemoryType(device, requirements.memoryTypeBits,
+                              vk::MemoryPropertyFlagBits::eHostVisible |
+                                  vk::MemoryPropertyFlagBits::eHostCoherent);
+        ASSERT_MSG(type, "No host visible and coherent memory type found");
+    }
+    const vk::MemoryAllocateInfo alloc_ci(requirements.size, *type);
+    memory = dev.allocateMemoryUnique(alloc_ci, nullptr, dld);
+
+    dev.bindBufferMemory(*buffer, *memory, 0, dld);
+}
+
+void VKStreamBuffer::ReserveWatches(std::vector<Watch>& watches, std::size_t grow_size) {
+    watches.resize(watches.size() + grow_size);
+}
+
+void VKStreamBuffer::WaitPendingOperations(u64 requested_upper_bound) {
+    if (!invalidation_mark) {
+        return;
+    }
+    while (requested_upper_bound < wait_bound && wait_cursor < *invalidation_mark) {
+        auto& watch = previous_watches[wait_cursor];
+        wait_bound = watch.upper_bound;
+        watch.fence.Wait();
+        ++wait_cursor;
+    }
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_stream_buffer.h

@@ -4,28 +4,24 @@
 
 #pragma once
 
-#include <memory>
 #include <optional>
 #include <tuple>
 #include <vector>
 
 #include "common/common_types.h"
 #include "video_core/renderer_vulkan/declarations.h"
-#include "video_core/renderer_vulkan/vk_memory_manager.h"
 
 namespace Vulkan {
 
 class VKDevice;
 class VKFence;
 class VKFenceWatch;
-class VKResourceManager;
 class VKScheduler;
 
-class VKStreamBuffer {
+class VKStreamBuffer final {
 public:
-    explicit VKStreamBuffer(const VKDevice& device, VKMemoryManager& memory_manager,
-                            VKScheduler& scheduler, u64 size, vk::BufferUsageFlags usage,
-                            vk::AccessFlags access, vk::PipelineStageFlags pipeline_stage);
+    explicit VKStreamBuffer(const VKDevice& device, VKScheduler& scheduler,
+                            vk::BufferUsageFlags usage);
     ~VKStreamBuffer();
 
     /**
@@ -34,39 +30,47 @@ public:
      * @returns A tuple in the following order: Raw memory pointer (with offset added), buffer
      * offset and a boolean that's true when buffer has been invalidated.
      */
-    std::tuple<u8*, u64, bool> Reserve(u64 size);
+    std::tuple<u8*, u64, bool> Map(u64 size, u64 alignment);
 
     /// Ensures that "size" bytes of memory are available to the GPU, potentially recording a copy.
-    void Send(u64 size);
+    void Unmap(u64 size);
 
-    vk::Buffer GetBuffer() const {
+    vk::Buffer GetHandle() const {
         return *buffer;
     }
 
 private:
+    struct Watch final {
+        VKFenceWatch fence;
+        u64 upper_bound{};
+    };
+
     /// Creates Vulkan buffer handles committing the required the required memory.
-    void CreateBuffers(VKMemoryManager& memory_manager, vk::BufferUsageFlags usage);
+    void CreateBuffers(vk::BufferUsageFlags usage);
 
     /// Increases the amount of watches available.
-    void ReserveWatches(std::size_t grow_size);
+    void ReserveWatches(std::vector<Watch>& watches, std::size_t grow_size);
+
+    void WaitPendingOperations(u64 requested_upper_bound);
 
     const VKDevice& device;                      ///< Vulkan device manager.
     VKScheduler& scheduler;                      ///< Command scheduler.
-    const u64 buffer_size;                       ///< Total size of the stream buffer.
     const vk::AccessFlags access;                ///< Access usage of this stream buffer.
     const vk::PipelineStageFlags pipeline_stage; ///< Pipeline usage of this stream buffer.
 
-    UniqueBuffer buffer;   ///< Mapped buffer.
-    VKMemoryCommit commit; ///< Memory commit.
-    u8* mapped_pointer{};  ///< Pointer to the host visible commit
+    UniqueBuffer buffer;       ///< Mapped buffer.
+    UniqueDeviceMemory memory; ///< Memory allocation.
 
     u64 offset{};      ///< Buffer iterator.
     u64 mapped_size{}; ///< Size reserved for the current copy.
 
-    std::vector<std::unique_ptr<VKFenceWatch>> watches; ///< Total watches
-    std::size_t used_watches{}; ///< Count of watches, reset on invalidation.
-    std::optional<std::size_t>
-        invalidation_mark{}; ///< Number of watches used in the current invalidation.
+    std::vector<Watch> current_watches;           ///< Watches recorded in the current iteration.
+    std::size_t current_watch_cursor{};           ///< Count of watches, reset on invalidation.
+    std::optional<std::size_t> invalidation_mark; ///< Number of watches used in the previous cycle.
+
+    std::vector<Watch> previous_watches; ///< Watches used in the previous iteration.
+    std::size_t wait_cursor{};           ///< Last watch being waited for completion.
+    u64 wait_bound{};                    ///< Highest offset being watched for completion.
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_swapchain.cpp

@@ -123,7 +123,7 @@ bool VKSwapchain::Present(vk::Semaphore render_semaphore, VKFence& fence) {
 
     ASSERT(fences[image_index] == nullptr);
     fences[image_index] = &fence;
-    frame_index = (frame_index + 1) % image_count;
+    frame_index = (frame_index + 1) % static_cast<u32>(image_count);
     return recreated;
 }
 

src/video_core/renderer_vulkan/vk_swapchain.h

@@ -40,19 +40,19 @@ public:
         return extent;
     }
 
-    u32 GetImageCount() const {
+    std::size_t GetImageCount() const {
         return image_count;
     }
 
-    u32 GetImageIndex() const {
+    std::size_t GetImageIndex() const {
         return image_index;
     }
 
-    vk::Image GetImageIndex(u32 index) const {
+    vk::Image GetImageIndex(std::size_t index) const {
         return images[index];
     }
 
-    vk::ImageView GetImageViewIndex(u32 index) const {
+    vk::ImageView GetImageViewIndex(std::size_t index) const {
         return *image_views[index];
     }
 
@@ -77,7 +77,7 @@ private:
 
     UniqueSwapchainKHR swapchain;
 
-    u32 image_count{};
+    std::size_t image_count{};
     std::vector<vk::Image> images;
     std::vector<UniqueImageView> image_views;
     std::vector<UniqueFramebuffer> framebuffers;