[TEST] DR Off

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt

@@ -594,7 +594,7 @@ abstract class SettingsItem(
                     IntSetting.ANDROID_PIPELINE_WORKERS,
                     titleId = R.string.pipeline_worker_cores,
                     descriptionId = R.string.pipeline_worker_cores_description,
-                    min = 4,
+                    min = 1,
                     max = 8,
                     units = "cores"
                 )

src/android/app/src/main/jni/android_settings.h

@@ -147,7 +147,7 @@ namespace AndroidSettings {
                                                       &show_performance_overlay};
 
 
-        Settings::Setting<s32> pipeline_worker_count{linkage, 4, "pipeline_worker_count",
+        Settings::Setting<s32> pipeline_worker_count{linkage, 2, "pipeline_worker_count",
                                  Settings::Category::Android,
                                  Settings::Specialization::Default,
                                  true,

src/core/arm/nce/arm_nce.cpp

@@ -157,6 +157,8 @@ bool ArmNce::HandleGuestAlignmentFault(GuestContext* guest_ctx, void* raw_info,
     return HandleFailedGuestFault(guest_ctx, raw_info, raw_context);
 }
 
+constexpr size_t NCE_WRITE_FAULT_CLUSTER_PAGES = 4;
+
 bool ArmNce::HandleGuestAccessFault(GuestContext* guest_ctx, void* raw_info, void* raw_context) {
     auto* info = static_cast<siginfo_t*>(raw_info);
 
@@ -165,7 +167,7 @@ bool ArmNce::HandleGuestAccessFault(GuestContext* guest_ctx, void* raw_info, voi
     const Common::ProcessAddress addr =
         (reinterpret_cast<u64>(info->si_addr) & ~Memory::YUZU_PAGEMASK);
     auto& memory = guest_ctx->parent->m_running_thread->GetOwnerProcess()->GetMemory();
-    if (memory.InvalidateNCE(addr, Memory::YUZU_PAGESIZE)) {
+    if (memory.InvalidateNCE(addr, Memory::YUZU_PAGESIZE * NCE_WRITE_FAULT_CLUSTER_PAGES)) {
         // We handled the access successfully and are returning to guest code.
         return true;
     }

src/core/device_memory_manager.h

@@ -126,6 +126,10 @@ public:
     // New batch API to update multiple ranges with a single lock acquisition.
     void UpdatePagesCachedBatch(std::span<const std::pair<DAddr, size_t>> ranges, s32 delta);
 
+    void UpdateTexturePagesCount(DAddr addr, size_t size, s32 delta);
+
+    [[nodiscard]] bool IsRegionTextureCached(DAddr addr, size_t size) const noexcept;
+
 private:
     struct TranslationEntry {
         DAddr guest_page{};
@@ -234,6 +238,7 @@ private:
         (1ULL << (device_virtual_bits - page_bits)) / subentries;
     using CachedPages = std::array<CounterEntry, num_counter_entries>;
     std::unique_ptr<CachedPages> cached_pages;
+    std::unique_ptr<CachedPages> texture_cached_pages;
     Common::RangeMutex counter_guard;
     std::mutex mapping_guard;
 

src/core/device_memory_manager.inc

@@ -177,6 +177,7 @@ DeviceMemoryManager<Traits>::DeviceMemoryManager(const DeviceMemory& device_memo
 {
     impl = std::make_unique<DeviceMemoryManagerAllocator<Traits>>();
     cached_pages = std::make_unique<CachedPages>();
+    texture_cached_pages = std::make_unique<CachedPages>();
 
     const size_t total_virtual = device_as_size >> Memory::YUZU_PAGEBITS;
     for (size_t i = 0; i < total_virtual; i++) {
@@ -625,6 +626,28 @@ void DeviceMemoryManager<Traits>::UpdatePagesCachedCount(DAddr addr, size_t size
     UpdatePagesCachedCountNoLock(addr, size, delta);
 }
 
+template <typename Traits>
+void DeviceMemoryManager<Traits>::UpdateTexturePagesCount(DAddr addr, size_t size, s32 delta) {
+    Common::ScopedRangeLock lk(counter_guard, addr, size);
+    const size_t page_end = Common::DivCeil(addr + size, Memory::YUZU_PAGESIZE);
+    for (size_t page = addr >> Memory::YUZU_PAGEBITS; page != page_end; ++page) {
+        CounterAtomicType& count = texture_cached_pages->at(page >> subentries_shift).Count(page);
+        count.fetch_add(static_cast<CounterType>(delta), std::memory_order_release);
+    }
+}
+
+template <typename Traits>
+bool DeviceMemoryManager<Traits>::IsRegionTextureCached(DAddr addr, size_t size) const noexcept {
+    const size_t page_end = Common::DivCeil(addr + size, Memory::YUZU_PAGESIZE);
+    for (size_t page = addr >> Memory::YUZU_PAGEBITS; page != page_end; ++page) {
+        if (texture_cached_pages->at(page >> subentries_shift).Count(page).load(
+                std::memory_order_acquire) != 0) {
+            return true;
+        }
+    }
+    return false;
+}
+
 template <typename Traits>
 void DeviceMemoryManager<Traits>::UpdatePagesCachedBatch(std::span<const std::pair<DAddr, size_t>> ranges, s32 delta) {
     if (ranges.empty()) {

src/dynarmic/src/dynarmic/backend/arm64/emit_arm64_a32.cpp

@@ -34,8 +34,7 @@ oaknut::Label EmitA32Cond(oaknut::CodeGenerator& code, EmitContext&, IR::Cond co
     return pass;
 }
 
-void EmitA32LeafTerminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::LeafTerminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step);
-void EmitA32Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::Terminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step);
+void EmitA32Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step);
 
 void EmitA32Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::ReturnToDispatch, IR::LocationDescriptor, bool) {
     EmitRelocation(code, ctx, LinkTarget::ReturnToDispatcher);
@@ -126,53 +125,31 @@ void EmitA32Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::Fa
 
 void EmitA32Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::If terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     oaknut::Label pass = EmitA32Cond(code, ctx, terminal.if_);
-    EmitA32LeafTerminal(code, ctx, terminal.else_, initial_location, is_single_step);
+    EmitA32Terminal(code, ctx, terminal.else_, initial_location, is_single_step);
     code.l(pass);
-    EmitA32LeafTerminal(code, ctx, terminal.then_, initial_location, is_single_step);
+    EmitA32Terminal(code, ctx, terminal.then_, initial_location, is_single_step);
 }
 
 void EmitA32Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::CheckBit terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     oaknut::Label fail;
     code.LDRB(Wscratch0, SP, offsetof(StackLayout, check_bit));
     code.CBZ(Wscratch0, fail);
-    EmitA32LeafTerminal(code, ctx, terminal.then_, initial_location, is_single_step);
+    EmitA32Terminal(code, ctx, terminal.then_, initial_location, is_single_step);
     code.l(fail);
-    EmitA32LeafTerminal(code, ctx, terminal.else_, initial_location, is_single_step);
+    EmitA32Terminal(code, ctx, terminal.else_, initial_location, is_single_step);
 }
 
 void EmitA32Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::CheckHalt terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     oaknut::Label fail;
     code.LDAR(Wscratch0, Xhalt);
     code.CBNZ(Wscratch0, fail);
-    EmitA32LeafTerminal(code, ctx, terminal.else_, initial_location, is_single_step);
+    EmitA32Terminal(code, ctx, terminal.else_, initial_location, is_single_step);
     code.l(fail);
     EmitRelocation(code, ctx, LinkTarget::ReturnToDispatcher);
 }
 
-void EmitA32LeafTerminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::LeafTerminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
-    if (auto const x = std::get_if<IR::Term::ReturnToDispatch>(&terminal))
-        return EmitA32Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::LinkBlock>(&terminal))
-        return EmitA32Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::LinkBlockFast>(&terminal))
-        return EmitA32Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::PopRSBHint>(&terminal))
-        return EmitA32Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::FastDispatchHint>(&terminal))
-        return EmitA32Terminal(code, ctx, *x, initial_location, is_single_step);
-    UNREACHABLE();
-}
-
-void EmitA32Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::Terminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
-    if (auto const x = std::get_if<IR::Term::LeafTerminal>(&terminal))
-        return EmitA32LeafTerminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::If>(&terminal))
-        return EmitA32Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::CheckBit>(&terminal))
-        return EmitA32Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::CheckHalt>(&terminal))
-        return EmitA32Terminal(code, ctx, *x, initial_location, is_single_step);
-    UNREACHABLE();
+void EmitA32Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
+    boost::apply_visitor([&](const auto& t) { EmitA32Terminal(code, ctx, t, initial_location, is_single_step); }, terminal);
 }
 
 void EmitA32Terminal(oaknut::CodeGenerator& code, EmitContext& ctx) {

src/dynarmic/src/dynarmic/backend/arm64/emit_arm64_a64.cpp

@@ -33,8 +33,7 @@ oaknut::Label EmitA64Cond(oaknut::CodeGenerator& code, EmitContext&, IR::Cond co
     return pass;
 }
 
-void EmitA64LeafTerminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::LeafTerminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step);
-void EmitA64Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::Terminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step);
+void EmitA64Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step);
 
 void EmitA64Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::ReturnToDispatch, IR::LocationDescriptor, bool) {
     EmitRelocation(code, ctx, LinkTarget::ReturnToDispatcher);
@@ -109,53 +108,31 @@ void EmitA64Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::Fa
 
 void EmitA64Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::If terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     oaknut::Label pass = EmitA64Cond(code, ctx, terminal.if_);
-    EmitA64LeafTerminal(code, ctx, terminal.else_, initial_location, is_single_step);
+    EmitA64Terminal(code, ctx, terminal.else_, initial_location, is_single_step);
     code.l(pass);
-    EmitA64LeafTerminal(code, ctx, terminal.then_, initial_location, is_single_step);
+    EmitA64Terminal(code, ctx, terminal.then_, initial_location, is_single_step);
 }
 
 void EmitA64Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::CheckBit terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     oaknut::Label fail;
     code.LDRB(Wscratch0, SP, offsetof(StackLayout, check_bit));
     code.CBZ(Wscratch0, fail);
-    EmitA64LeafTerminal(code, ctx, terminal.then_, initial_location, is_single_step);
+    EmitA64Terminal(code, ctx, terminal.then_, initial_location, is_single_step);
     code.l(fail);
-    EmitA64LeafTerminal(code, ctx, terminal.else_, initial_location, is_single_step);
+    EmitA64Terminal(code, ctx, terminal.else_, initial_location, is_single_step);
 }
 
 void EmitA64Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::CheckHalt terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     oaknut::Label fail;
     code.LDAR(Wscratch0, Xhalt);
     code.CBNZ(Wscratch0, fail);
-    EmitA64LeafTerminal(code, ctx, terminal.else_, initial_location, is_single_step);
+    EmitA64Terminal(code, ctx, terminal.else_, initial_location, is_single_step);
     code.l(fail);
     EmitRelocation(code, ctx, LinkTarget::ReturnToDispatcher);
 }
 
-void EmitA64LeafTerminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::LeafTerminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
-    if (auto const x = std::get_if<IR::Term::ReturnToDispatch>(&terminal))
-        return EmitA64Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::LinkBlock>(&terminal))
-        return EmitA64Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::LinkBlockFast>(&terminal))
-        return EmitA64Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::PopRSBHint>(&terminal))
-        return EmitA64Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::FastDispatchHint>(&terminal))
-        return EmitA64Terminal(code, ctx, *x, initial_location, is_single_step);
-    UNREACHABLE();
-}
-
-void EmitA64Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::Terminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
-    if (auto const x = std::get_if<IR::Term::LeafTerminal>(&terminal))
-        return EmitA64LeafTerminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::If>(&terminal))
-        return EmitA64Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::CheckBit>(&terminal))
-        return EmitA64Terminal(code, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::CheckHalt>(&terminal))
-        return EmitA64Terminal(code, ctx, *x, initial_location, is_single_step);
-    UNREACHABLE();
+void EmitA64Terminal(oaknut::CodeGenerator& code, EmitContext& ctx, IR::Term::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
+    boost::apply_visitor([&](const auto& t) { EmitA64Terminal(code, ctx, t, initial_location, is_single_step); }, terminal);
 }
 
 void EmitA64Terminal(oaknut::CodeGenerator& code, EmitContext& ctx) {

src/dynarmic/src/dynarmic/backend/riscv64/emit_riscv64_a32.cpp

@@ -112,7 +112,6 @@ void EmitA32Cond(biscuit::Assembler& as, EmitContext&, IR::Cond cond, biscuit::L
     }
 }
 
-void EmitA32LeafTerminal(biscuit::Assembler& as, EmitContext& ctx, IR::Term::LeafTerminal terminal, IR::LocationDescriptor initial_location, bool is_single_step);
 void EmitA32Terminal(biscuit::Assembler& as, EmitContext& ctx, IR::Term::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step);
 
 void EmitA32Terminal(biscuit::Assembler& as, EmitContext& ctx, IR::Term::ReturnToDispatch, IR::LocationDescriptor, bool) {
@@ -171,18 +170,18 @@ void EmitA32Terminal(biscuit::Assembler& as, EmitContext& ctx, IR::Term::FastDis
 void EmitA32Terminal(biscuit::Assembler& as, EmitContext& ctx, IR::Term::If terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     biscuit::Label pass;
     EmitA32Cond(as, ctx, terminal.if_, &pass);
-    EmitA32LeafTerminal(as, ctx, terminal.else_, initial_location, is_single_step);
+    EmitA32Terminal(as, ctx, terminal.else_, initial_location, is_single_step);
     as.Bind(&pass);
-    EmitA32LeafTerminal(as, ctx, terminal.then_, initial_location, is_single_step);
+    EmitA32Terminal(as, ctx, terminal.then_, initial_location, is_single_step);
 }
 
 void EmitA32Terminal(biscuit::Assembler& as, EmitContext& ctx, IR::Term::CheckBit terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     biscuit::Label fail;
     as.LBU(Xscratch0, offsetof(StackLayout, check_bit), Xstate);
     as.BEQZ(Xscratch0, &fail);
-    EmitA32LeafTerminal(as, ctx, terminal.then_, initial_location, is_single_step);
+    EmitA32Terminal(as, ctx, terminal.then_, initial_location, is_single_step);
     as.Bind(&fail);
-    EmitA32LeafTerminal(as, ctx, terminal.else_, initial_location, is_single_step);
+    EmitA32Terminal(as, ctx, terminal.else_, initial_location, is_single_step);
 }
 
 void EmitA32Terminal(biscuit::Assembler& as, EmitContext& ctx, IR::Term::CheckHalt terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
@@ -190,35 +189,13 @@ void EmitA32Terminal(biscuit::Assembler& as, EmitContext& ctx, IR::Term::CheckHa
     as.LWU(Xscratch0, 0, Xhalt);
     as.FENCE(biscuit::FenceOrder::RW, biscuit::FenceOrder::RW);
     as.BNEZ(Xscratch0, &fail);
-    EmitA32LeafTerminal(as, ctx, terminal.else_, initial_location, is_single_step);
+    EmitA32Terminal(as, ctx, terminal.else_, initial_location, is_single_step);
     as.Bind(&fail);
     EmitRelocation(as, ctx, LinkTarget::ReturnFromRunCode);
 }
 
-void EmitA32LeafTerminal(biscuit::Assembler& as, EmitContext& ctx, IR::Term::LeafTerminal terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
-    if (auto const x = std::get_if<IR::Term::ReturnToDispatch>(&terminal))
-        return EmitA32LeafTerminal(as, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::LinkBlock>(&terminal))
-        return EmitA32LeafTerminal(as, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::LinkBlockFast>(&terminal))
-        return EmitA32LeafTerminal(as, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::PopRSBHint>(&terminal))
-        return EmitA32LeafTerminal(as, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::FastDispatchHint>(&terminal))
-        return EmitA32LeafTerminal(as, ctx, *x, initial_location, is_single_step);
-    UNREACHABLE();
-}
-
 void EmitA32Terminal(biscuit::Assembler& as, EmitContext& ctx, IR::Term::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
-    if (auto const x = std::get_if<IR::Term::LeafTerminal>(&terminal))
-        return EmitA32LeafTerminal(as, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::If>(&terminal))
-        return EmitA32Terminal(as, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::CheckBit>(&terminal))
-        return EmitA32Terminal(as, ctx, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::CheckHalt>(&terminal))
-        return EmitA32Terminal(as, ctx, *x, initial_location, is_single_step);
-    UNREACHABLE();
+    boost::apply_visitor([&](const auto& t) { EmitA32Terminal(as, ctx, t, initial_location, is_single_step); }, terminal);
 }
 
 void EmitA32Terminal(biscuit::Assembler& as, EmitContext& ctx) {

src/dynarmic/src/dynarmic/backend/x64/a32_emit_x64.cpp

@@ -175,7 +175,7 @@ finish_this_inst:
     if (conf.enable_cycle_counting)
         EmitAddCycles(block.CycleCount());
     code.mov(rbp, code.qword[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, abi_base_pointer)]);
-    EmitTerminal(block.terminal, ctx.Location().SetSingleStepping(false), ctx.IsSingleStep());
+    EmitTerminal(block.GetTerminal(), ctx.Location().SetSingleStepping(false), ctx.IsSingleStep());
     code.int3();
 
     for (auto& deferred_emit : ctx.deferred_emits)
@@ -219,7 +219,7 @@ void A32EmitX64::EmitCondPrelude(const A32EmitContext& ctx) {
     if (conf.enable_cycle_counting) {
         EmitAddCycles(ctx.block.ConditionFailedCycleCount());
     }
-    EmitLeafTerminal(IR::Term::LinkBlock{ctx.block.ConditionFailedLocation()}, ctx.Location().SetSingleStepping(false), ctx.IsSingleStep());
+    EmitTerminal(IR::Term::LinkBlock{ctx.block.ConditionFailedLocation()}, ctx.Location().SetSingleStepping(false), ctx.IsSingleStep());
     code.L(pass);
 }
 
@@ -1155,12 +1155,11 @@ void A32EmitX64::EmitSetUpperLocationDescriptor(IR::LocationDescriptor new_locat
 }
 
 namespace {
-bool EmitTerminalImpl(A32EmitX64& e, IR::Term::ReturnToDispatch, IR::LocationDescriptor, bool) {
+void EmitTerminalImpl(A32EmitX64& e, IR::Term::ReturnToDispatch, IR::LocationDescriptor, bool) {
     e.code.ReturnFromRunCode();
-    return true;
 }
 
-bool EmitTerminalImpl(A32EmitX64& e, IR::Term::LinkBlock terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
+void EmitTerminalImpl(A32EmitX64& e, IR::Term::LinkBlock terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     e.EmitSetUpperLocationDescriptor(terminal.next, initial_location);
     if (!e.conf.HasOptimization(OptimizationFlag::BlockLinking) || is_single_step) {
         e.code.mov(MJitStateReg(A32::Reg::PC), A32::LocationDescriptor{terminal.next}.PC());
@@ -1187,10 +1186,9 @@ bool EmitTerminalImpl(A32EmitX64& e, IR::Term::LinkBlock terminal, IR::LocationD
         e.PushRSBHelper(rax, rbx, terminal.next);
         e.code.ForceReturnFromRunCode();
     }
-    return true;
 }
 
-bool EmitTerminalImpl(A32EmitX64& e, IR::Term::LinkBlockFast terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
+void EmitTerminalImpl(A32EmitX64& e, IR::Term::LinkBlockFast terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     e.EmitSetUpperLocationDescriptor(terminal.next, initial_location);
     if (!e.conf.HasOptimization(OptimizationFlag::BlockLinking) || is_single_step) {
         e.code.mov(MJitStateReg(A32::Reg::PC), A32::LocationDescriptor{terminal.next}.PC());
@@ -1203,78 +1201,55 @@ bool EmitTerminalImpl(A32EmitX64& e, IR::Term::LinkBlockFast terminal, IR::Locat
             e.EmitPatchJmp(terminal.next);
         }
     }
-    return true;
 }
 
-bool EmitTerminalImpl(A32EmitX64& e, IR::Term::PopRSBHint, IR::LocationDescriptor, bool is_single_step) {
+void EmitTerminalImpl(A32EmitX64& e, IR::Term::PopRSBHint, IR::LocationDescriptor, bool is_single_step) {
     if (!e.conf.HasOptimization(OptimizationFlag::ReturnStackBuffer) || is_single_step) {
         e.code.ReturnFromRunCode();
     } else {
         e.code.jmp(e.terminal_handler_pop_rsb_hint);
     }
-    return true;
 }
 
-bool EmitTerminalImpl(A32EmitX64& e, IR::Term::FastDispatchHint, IR::LocationDescriptor, bool is_single_step) {
+void EmitTerminalImpl(A32EmitX64& e, IR::Term::FastDispatchHint, IR::LocationDescriptor, bool is_single_step) {
     if (!e.conf.HasOptimization(OptimizationFlag::FastDispatch) || is_single_step) {
         e.code.ReturnFromRunCode();
     } else {
         e.code.jmp(e.terminal_handler_fast_dispatch_hint);
     }
-    return true;
 }
 
-bool EmitTerminalImpl(A32EmitX64& e, IR::Term::If terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
+void EmitTerminalImpl(A32EmitX64& e, IR::Term::If terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     Xbyak::Label pass = e.EmitCond(terminal.if_);
-    e.EmitLeafTerminal(terminal.else_, initial_location, is_single_step);
+    e.EmitTerminal(terminal.else_, initial_location, is_single_step);
     e.code.L(pass);
-    e.EmitLeafTerminal(terminal.then_, initial_location, is_single_step);
-    return true;
+    e.EmitTerminal(terminal.then_, initial_location, is_single_step);
 }
 
-bool EmitTerminalImpl(A32EmitX64& e, IR::Term::CheckBit terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
+void EmitTerminalImpl(A32EmitX64& e, IR::Term::CheckBit terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     Xbyak::Label fail;
     e.code.cmp(e.code.byte[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, check_bit)], u8(0));
     e.code.jz(fail);
-    e.EmitLeafTerminal(terminal.then_, initial_location, is_single_step);
+    e.EmitTerminal(terminal.then_, initial_location, is_single_step);
     e.code.L(fail);
-    e.EmitLeafTerminal(terminal.else_, initial_location, is_single_step);
-    return true;
+    e.EmitTerminal(terminal.else_, initial_location, is_single_step);
 }
 
-bool EmitTerminalImpl(A32EmitX64& e, IR::Term::CheckHalt terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
+void EmitTerminalImpl(A32EmitX64& e, IR::Term::CheckHalt terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     e.code.cmp(dword[e.code.ABI_JIT_PTR + offsetof(A32JitState, halt_reason)], 0);
     e.code.jne(e.code.GetForceReturnFromRunCodeAddress());
-    e.EmitLeafTerminal(terminal.else_, initial_location, is_single_step);
-    return true;
+    e.EmitTerminal(terminal.else_, initial_location, is_single_step);
 }
 
-}
-
-bool A32EmitX64::EmitLeafTerminal(IR::Term::LeafTerminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept {
-    if (auto const x = std::get_if<IR::Term::ReturnToDispatch>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::LinkBlock>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::LinkBlockFast>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::PopRSBHint>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::FastDispatchHint>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
+void EmitTerminalImpl(A32EmitX64&, IR::Term::Invalid, IR::LocationDescriptor, bool) {
     UNREACHABLE();
 }
+}
 
-bool A32EmitX64::EmitTerminal(IR::Term::Terminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept {
-    if (auto const e = std::get_if<IR::Term::LeafTerminal>(&terminal))
-        return EmitLeafTerminal(*e, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::If>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::CheckBit>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::CheckHalt>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    UNREACHABLE();
+void A32EmitX64::EmitTerminal(IR::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept {
+    boost::apply_visitor([this, initial_location, is_single_step](auto x) {
+        EmitTerminalImpl(*this, x, initial_location, is_single_step);
+    }, terminal);
 }
 
 void A32EmitX64::EmitPatchJg(const IR::LocationDescriptor& target_desc, CodePtr target_code_ptr) {

src/dynarmic/src/dynarmic/backend/x64/a32_emit_x64.h

@@ -112,8 +112,7 @@ public:
 
     // Terminal instruction emitters
     void EmitSetUpperLocationDescriptor(IR::LocationDescriptor new_location, IR::LocationDescriptor old_location);
-    bool EmitLeafTerminal(IR::Term::LeafTerminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept override;
-    bool EmitTerminal(IR::Term::Terminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept override;
+    void EmitTerminal(IR::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept override;
 
     // Patching
     void Unpatch(const IR::LocationDescriptor& target_desc) override;

src/dynarmic/src/dynarmic/backend/x64/a64_emit_x64.cpp

@@ -12,7 +12,6 @@
 #include <fmt/ostream.h>
 #include "common/assert.h"
 #include "common/common_types.h"
-#include "dynarmic/ir/terminal.h"
 #include "dynarmic/mcl/integer_of_size.hpp"
 #include <boost/container/static_vector.hpp>
 
@@ -148,7 +147,7 @@ finish_this_inst:
     if (conf.enable_cycle_counting)
         EmitAddCycles(block.CycleCount());
     code.mov(rbp, code.qword[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, abi_base_pointer)]);
-    EmitTerminal(block.terminal, ctx.Location().SetSingleStepping(false), ctx.IsSingleStep());
+    EmitTerminal(block.GetTerminal(), ctx.Location().SetSingleStepping(false), ctx.IsSingleStep());
     code.int3();
     for (auto& deferred_emit : ctx.deferred_emits)
         deferred_emit();
@@ -618,12 +617,11 @@ std::string A64EmitX64::LocationDescriptorToFriendlyName(const IR::LocationDescr
 }
 
 namespace {
-bool EmitTerminalImpl(A64EmitX64& e, IR::Term::ReturnToDispatch, IR::LocationDescriptor, bool) {
+void EmitTerminalImpl(A64EmitX64& e, IR::Term::ReturnToDispatch, IR::LocationDescriptor, bool) {
     e.code.ReturnFromRunCode();
-    return true;
 }
 
-bool EmitTerminalImpl(A64EmitX64& e, IR::Term::LinkBlock terminal, IR::LocationDescriptor, bool is_single_step) {
+void EmitTerminalImpl(A64EmitX64& e, IR::Term::LinkBlock terminal, IR::LocationDescriptor, bool is_single_step) {
     // Used for patches and linking
     if (e.conf.HasOptimization(OptimizationFlag::BlockLinking) && !is_single_step) {
         if (e.conf.enable_cycle_counting) {
@@ -651,10 +649,9 @@ bool EmitTerminalImpl(A64EmitX64& e, IR::Term::LinkBlock terminal, IR::LocationD
         e.code.mov(qword[e.code.ABI_JIT_PTR + offsetof(A64JitState, pc)], rax);
         e.code.ReturnFromRunCode();
     }
-    return true;
 }
 
-bool EmitTerminalImpl(A64EmitX64& e, IR::Term::LinkBlockFast terminal, IR::LocationDescriptor, bool is_single_step) {
+void EmitTerminalImpl(A64EmitX64& e, IR::Term::LinkBlockFast terminal, IR::LocationDescriptor, bool is_single_step) {
     if (e.conf.HasOptimization(OptimizationFlag::BlockLinking) && !is_single_step) {
         e.patch_information[terminal.next].jmp.push_back(e.code.getCurr());
         if (auto next_bb = e.GetBasicBlock(terminal.next)) {
@@ -667,86 +664,63 @@ bool EmitTerminalImpl(A64EmitX64& e, IR::Term::LinkBlockFast terminal, IR::Locat
         e.code.mov(qword[e.code.ABI_JIT_PTR + offsetof(A64JitState, pc)], rax);
         e.code.ReturnFromRunCode();
     }
-    return true;
 }
 
-bool EmitTerminalImpl(A64EmitX64& e, IR::Term::PopRSBHint, IR::LocationDescriptor, bool is_single_step) {
+void EmitTerminalImpl(A64EmitX64& e, IR::Term::PopRSBHint, IR::LocationDescriptor, bool is_single_step) {
     if (e.conf.HasOptimization(OptimizationFlag::ReturnStackBuffer) && !is_single_step) {
         e.code.jmp(e.terminal_handler_pop_rsb_hint);
     } else {
         e.code.ReturnFromRunCode();
     }
-    return true;
 }
 
-bool EmitTerminalImpl(A64EmitX64& e, IR::Term::FastDispatchHint, IR::LocationDescriptor, bool is_single_step) {
+void EmitTerminalImpl(A64EmitX64& e, IR::Term::FastDispatchHint, IR::LocationDescriptor, bool is_single_step) {
     if (!e.conf.HasOptimization(OptimizationFlag::FastDispatch) || is_single_step) {
         e.code.ReturnFromRunCode();
     } else {
         e.code.jmp(e.terminal_handler_fast_dispatch_hint);
     }
-    return true;
 }
 
-bool EmitTerminalImpl(A64EmitX64& e, IR::Term::If terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
+void EmitTerminalImpl(A64EmitX64& e, IR::Term::If terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     switch (terminal.if_) {
     case IR::Cond::AL:
     case IR::Cond::NV:
-        e.EmitLeafTerminal(terminal.then_, initial_location, is_single_step);
+        e.EmitTerminal(terminal.then_, initial_location, is_single_step);
         break;
     default:
         Xbyak::Label pass = e.EmitCond(terminal.if_);
-        e.EmitLeafTerminal(terminal.else_, initial_location, is_single_step);
+        e.EmitTerminal(terminal.else_, initial_location, is_single_step);
         e.code.L(pass);
-        e.EmitLeafTerminal(terminal.then_, initial_location, is_single_step);
+        e.EmitTerminal(terminal.then_, initial_location, is_single_step);
         break;
     }
-    return true;
 }
 
-bool EmitTerminalImpl(A64EmitX64& e, IR::Term::CheckBit terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
+void EmitTerminalImpl(A64EmitX64& e, IR::Term::CheckBit terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     Xbyak::Label fail;
     e.code.cmp(e.code.byte[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, check_bit)], u8(0));
     e.code.jz(fail);
-    e.EmitLeafTerminal(terminal.then_, initial_location, is_single_step);
+    e.EmitTerminal(terminal.then_, initial_location, is_single_step);
     e.code.L(fail);
-    e.EmitLeafTerminal(terminal.else_, initial_location, is_single_step);
-    return true;
+    e.EmitTerminal(terminal.else_, initial_location, is_single_step);
 }
 
-bool EmitTerminalImpl(A64EmitX64& e, IR::Term::CheckHalt terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
+void EmitTerminalImpl(A64EmitX64& e, IR::Term::CheckHalt terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
     e.code.cmp(dword[e.code.ABI_JIT_PTR + offsetof(A64JitState, halt_reason)], 0);
     e.code.jne(e.code.GetForceReturnFromRunCodeAddress());
-    e.EmitLeafTerminal(terminal.else_, initial_location, is_single_step);
-    return true;
+    e.EmitTerminal(terminal.else_, initial_location, is_single_step);
 }
 
-}
-
-bool A64EmitX64::EmitLeafTerminal(IR::Term::LeafTerminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept {
-    if (auto const x = std::get_if<IR::Term::ReturnToDispatch>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::LinkBlock>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::LinkBlockFast>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::PopRSBHint>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::FastDispatchHint>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
+void EmitTerminalImpl(A64EmitX64&, IR::Term::Invalid, IR::LocationDescriptor, bool) {
     UNREACHABLE();
 }
+}
 
-bool A64EmitX64::EmitTerminal(IR::Term::Terminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept {
-    if (auto const x = std::get_if<IR::Term::LeafTerminal>(&terminal))
-        return EmitLeafTerminal(*x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::If>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::CheckBit>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    if (auto const x = std::get_if<IR::Term::CheckHalt>(&terminal))
-        return EmitTerminalImpl(*this, *x, initial_location, is_single_step);
-    UNREACHABLE();
+void A64EmitX64::EmitTerminal(IR::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept {
+    boost::apply_visitor([this, initial_location, is_single_step](auto x) {
+        EmitTerminalImpl(*this, x, initial_location, is_single_step);
+    }, terminal);
 }
 
 void A64EmitX64::EmitPatchJg(const IR::LocationDescriptor& target_desc, CodePtr target_code_ptr) {

src/dynarmic/src/dynarmic/backend/x64/a64_emit_x64.h

@@ -107,8 +107,7 @@ public:
     void EmitExclusiveWriteMemoryInline(A64EmitContext& ctx, IR::Inst* inst);
 
     // Terminal instruction emitters
-    bool EmitLeafTerminal(IR::Term::LeafTerminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept override;
-    bool EmitTerminal(IR::Term::Terminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept override;
+    void EmitTerminal(IR::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept override;
 
     // Patching
     void Unpatch(const IR::LocationDescriptor& target_desc) override;

src/dynarmic/src/dynarmic/backend/x64/emit_x64.h

@@ -111,8 +111,7 @@ public:
 #ifndef NDEBUG
     void EmitVerboseDebuggingOutput(RegAlloc& reg_alloc);
 #endif
-    virtual bool EmitLeafTerminal(IR::Term::LeafTerminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept = 0;
-    virtual bool EmitTerminal(IR::Term::Terminal const& terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept = 0;
+    virtual void EmitTerminal(IR::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step) noexcept = 0;
 
     // Patching
     struct PatchInformation {

src/dynarmic/src/dynarmic/ir/basic_block.cpp

@@ -66,44 +66,43 @@ void Block::Reset(LocationDescriptor location_) noexcept {
     location = location_;
     end_location = location_;
     cond = Cond::AL;
-    terminal = std::monostate{};
+    terminal = Term::Invalid{};
     cond_failed_cycle_count = 0;
     cycle_count = 0;
     ASSERT(instructions.size() == 0);
 }
 
-static std::string TerminalToString(const Term::Terminal& terminal_variant) noexcept {
-    // struct : boost::static_visitor<std::string> {
-    //     std::string operator()(const std::monostate&) const {
-    //         return "<invalid>";
-    //     }
-    //     std::string operator()(const Term::ReturnToDispatch&) const {
-    //         return "ReturnToDispatch{}";
-    //     }
-    //     std::string operator()(const Term::LinkBlock& terminal) const {
-    //         return fmt::format("LinkBlock{{{}}}", terminal.next);
-    //     }
-    //     std::string operator()(const Term::LinkBlockFast& terminal) const {
-    //         return fmt::format("LinkBlockFast{{{}}}", terminal.next);
-    //     }
-    //     std::string operator()(const Term::PopRSBHint&) const {
-    //         return "PopRSBHint{}";
-    //     }
-    //     std::string operator()(const Term::FastDispatchHint&) const {
-    //         return "FastDispatchHint{}";
-    //     }
-    //     std::string operator()(const Term::If& terminal) const {
-    //         return fmt::format("If{{{}, {}, {}}}", A64::CondToString(terminal.if_), TerminalToString(terminal.then_), TerminalToString(terminal.else_));
-    //     }
-    //     std::string operator()(const Term::CheckBit& terminal) const {
-    //         return fmt::format("CheckBit{{{}, {}}}", TerminalToString(terminal.then_), TerminalToString(terminal.else_));
-    //     }
-    //     std::string operator()(const Term::CheckHalt& terminal) const {
-    //         return fmt::format("CheckHalt{{{}}}", TerminalToString(terminal.else_));
-    //     }
-    // } visitor;
-    // return boost::apply_visitor(visitor, terminal_variant);
-    return "";
+static std::string TerminalToString(const Terminal& terminal_variant) noexcept {
+    struct : boost::static_visitor<std::string> {
+        std::string operator()(const Term::Invalid&) const {
+            return "<invalid terminal>";
+        }
+        std::string operator()(const Term::ReturnToDispatch&) const {
+            return "ReturnToDispatch{}";
+        }
+        std::string operator()(const Term::LinkBlock& terminal) const {
+            return fmt::format("LinkBlock{{{}}}", terminal.next);
+        }
+        std::string operator()(const Term::LinkBlockFast& terminal) const {
+            return fmt::format("LinkBlockFast{{{}}}", terminal.next);
+        }
+        std::string operator()(const Term::PopRSBHint&) const {
+            return "PopRSBHint{}";
+        }
+        std::string operator()(const Term::FastDispatchHint&) const {
+            return "FastDispatchHint{}";
+        }
+        std::string operator()(const Term::If& terminal) const {
+            return fmt::format("If{{{}, {}, {}}}", A64::CondToString(terminal.if_), TerminalToString(terminal.then_), TerminalToString(terminal.else_));
+        }
+        std::string operator()(const Term::CheckBit& terminal) const {
+            return fmt::format("CheckBit{{{}, {}}}", TerminalToString(terminal.then_), TerminalToString(terminal.else_));
+        }
+        std::string operator()(const Term::CheckHalt& terminal) const {
+            return fmt::format("CheckHalt{{{}}}", TerminalToString(terminal.else_));
+        }
+    } visitor;
+    return boost::apply_visitor(visitor, terminal_variant);
 }
 
 std::string DumpBlock(const IR::Block& block) noexcept {

src/dynarmic/src/dynarmic/ir/basic_block.h

@@ -114,22 +114,22 @@ public:
     }
 
     /// Gets the terminal instruction for this basic block.
-    inline Term::Terminal GetTerminal() const noexcept {
+    inline Terminal GetTerminal() const noexcept {
         return terminal;
     }
     /// Sets the terminal instruction for this basic block.
-    inline void SetTerminal(Term::Terminal term) noexcept {
+    inline void SetTerminal(Terminal term) noexcept {
         ASSERT(!HasTerminal() && "Terminal has already been set.");
         terminal = std::move(term);
     }
     /// Replaces the terminal instruction for this basic block.
-    inline void ReplaceTerminal(Term::Terminal term) noexcept {
+    inline void ReplaceTerminal(Terminal term) noexcept {
         ASSERT(HasTerminal() && "Terminal has not been set.");
         terminal = std::move(term);
     }
     /// Determines whether or not this basic block has a terminal instruction.
     inline bool HasTerminal() const noexcept {
-        return !std::holds_alternative<std::monostate>(terminal);
+        return terminal.which() != 0;
     }
 
     /// Gets a mutable reference to the cycle count for this basic block.
@@ -156,7 +156,7 @@ public:
     /// Conditional to pass in order to execute this block
     Cond cond = Cond::AL;
     /// Terminal instruction of this block.
-    Term::Terminal terminal = std::monostate{};
+    Terminal terminal = Term::Invalid{};
     /// Number of cycles this block takes to execute if the conditional fails.
     size_t cond_failed_cycle_count = 0;
     /// Number of cycles this block takes to execute.

src/dynarmic/src/dynarmic/ir/ir_emitter.h

@@ -2943,7 +2943,7 @@ public:
         Inst(Opcode::CallHostFunction, Imm64(std::bit_cast<u64>(fn)), arg1, arg2, arg3);
     }
 
-    void SetTerm(const Term::Terminal& terminal) {
+    void SetTerm(const Terminal& terminal) {
         block.SetTerminal(terminal);
     }
 

src/dynarmic/src/dynarmic/ir/terminal.h

@@ -8,7 +8,7 @@
 
 #pragma once
 
-#include <variant>
+#include <boost/variant.hpp>
 #include "common/common_types.h"
 
 #include "dynarmic/ir/cond.h"
@@ -17,89 +17,106 @@
 namespace Dynarmic::IR {
 namespace Term {
 
-/// This terminal instruction returns control to the dispatcher.
-/// The dispatcher will use the current cpu state to determine what comes next.
+struct Invalid {};
+
+/**
+ * This terminal instruction returns control to the dispatcher.
+ * The dispatcher will use the current cpu state to determine what comes next.
+ */
 struct ReturnToDispatch {};
 
-/// This terminal instruction jumps to the basic block described by `next` if we have enough
-/// cycles remaining. If we do not have enough cycles remaining, we return to the
-/// dispatcher, which will return control to the host.
+/**
+ * This terminal instruction jumps to the basic block described by `next` if we have enough
+ * cycles remaining. If we do not have enough cycles remaining, we return to the
+ * dispatcher, which will return control to the host.
+ */
 struct LinkBlock {
-    explicit LinkBlock(const LocationDescriptor& next_) : next(next_) {}
+    explicit LinkBlock(const LocationDescriptor& next_)
+            : next(next_) {}
     LocationDescriptor next;  ///< Location descriptor for next block.
 };
 
-/// This terminal instruction jumps to the basic block described by `next` unconditionally.
-/// This is an optimization and MUST only be emitted when this is guaranteed not to result
-/// in hanging, even in the face of other optimizations. (In practice, this means that only
-/// forward jumps to short-ish blocks would use this instruction.)
-/// A backend that doesn't support this optimization may choose to implement this exactly
-/// as LinkBlock.
+/**
+ * This terminal instruction jumps to the basic block described by `next` unconditionally.
+ * This is an optimization and MUST only be emitted when this is guaranteed not to result
+ * in hanging, even in the face of other optimizations. (In practice, this means that only
+ * forward jumps to short-ish blocks would use this instruction.)
+ * A backend that doesn't support this optimization may choose to implement this exactly
+ * as LinkBlock.
+ */
 struct LinkBlockFast {
-    explicit LinkBlockFast(const LocationDescriptor& next_) : next(next_) {}
+    explicit LinkBlockFast(const LocationDescriptor& next_)
+            : next(next_) {}
     LocationDescriptor next;  ///< Location descriptor for next block.
 };
 
-/// This terminal instruction checks the top of the Return Stack Buffer against the current
-/// location descriptor. If RSB lookup fails, control is returned to the dispatcher.
-/// This is an optimization for faster function calls. A backend that doesn't support
-/// this optimization or doesn't have a RSB may choose to implement this exactly as
-/// ReturnToDispatch.
+/**
+ * This terminal instruction checks the top of the Return Stack Buffer against the current
+ * location descriptor. If RSB lookup fails, control is returned to the dispatcher.
+ * This is an optimization for faster function calls. A backend that doesn't support
+ * this optimization or doesn't have a RSB may choose to implement this exactly as
+ * ReturnToDispatch.
+ */
 struct PopRSBHint {};
 
-/// This terminal instruction performs a lookup of the current location descriptor in the
-/// fast dispatch lookup table. A backend that doesn't support this optimization may choose
-/// to implement this exactly as ReturnToDispatch.
+/**
+ * This terminal instruction performs a lookup of the current location descriptor in the
+ * fast dispatch lookup table. A backend that doesn't support this optimization may choose
+ * to implement this exactly as ReturnToDispatch.
+ */
 struct FastDispatchHint {};
 
 struct If;
 struct CheckBit;
 struct CheckHalt;
-
-/// Non recursive kind of terminal
-using LeafTerminal = std::variant<
-    std::monostate,
+/// A Terminal is the terminal instruction in a MicroBlock.
+using Terminal = boost::variant<
+    Invalid,
     ReturnToDispatch,
     LinkBlock,
     LinkBlockFast,
     PopRSBHint,
-    FastDispatchHint
->;
+    FastDispatchHint,
+    boost::recursive_wrapper<If>,
+    boost::recursive_wrapper<CheckBit>,
+    boost::recursive_wrapper<CheckHalt>>;
 
-/// A Terminal is the terminal instruction in a MicroBlock.
-using Terminal = std::variant<
-    std::monostate,
-    LeafTerminal,
-    If,
-    CheckBit,
-    CheckHalt
->;
-
-/// This terminal instruction conditionally executes one terminal or another depending
-/// on the run-time state of the ARM flags.
+/**
+ * This terminal instruction conditionally executes one terminal or another depending
+ * on the run-time state of the ARM flags.
+ */
 struct If {
-    explicit If(Cond if_, LeafTerminal then_, LeafTerminal else_) : if_(if_), then_(std::move(then_)), else_(std::move(else_)) {}
+    If(Cond if_, Terminal then_, Terminal else_)
+            : if_(if_), then_(std::move(then_)), else_(std::move(else_)) {}
     Cond if_;
-    LeafTerminal then_;
-    LeafTerminal else_;
+    Terminal then_;
+    Terminal else_;
 };
 
-/// This terminal instruction conditionally executes one terminal or another depending
-/// on the run-time state of the check bit.
-/// then_ is executed if the check bit is non-zero, otherwise else_ is executed.
+/**
+ * This terminal instruction conditionally executes one terminal or another depending
+ * on the run-time state of the check bit.
+ * then_ is executed if the check bit is non-zero, otherwise else_ is executed.
+ */
 struct CheckBit {
-    explicit CheckBit(LeafTerminal then_, LeafTerminal else_) : then_(std::move(then_)), else_(std::move(else_)) {}
-    LeafTerminal then_;
-    LeafTerminal else_;
+    CheckBit(Terminal then_, Terminal else_)
+            : then_(std::move(then_)), else_(std::move(else_)) {}
+    Terminal then_;
+    Terminal else_;
 };
 
-/// This terminal instruction checks if a halt was requested. If it wasn't, else_ is
-/// executed.
+/**
+ * This terminal instruction checks if a halt was requested. If it wasn't, else_ is
+ * executed.
+ */
 struct CheckHalt {
-    explicit CheckHalt(LeafTerminal else_) : else_(std::move(else_)) {}
-    LeafTerminal else_;
+    explicit CheckHalt(Terminal else_)
+            : else_(std::move(else_)) {}
+    Terminal else_;
 };
 
 }  // namespace Term
 
+using Term::Terminal;
+
 }  // namespace Dynarmic::IR

src/dynarmic/tests/A32/fuzz_arm.cpp

@@ -43,20 +43,32 @@ namespace {
 using namespace Dynarmic;
 
 template<typename Fn>
-bool AnyLocationDescriptorForTerminalHas(IR::Term::Terminal terminal, Fn fn) {
-    if (auto const e = std::get_if<IR::Term::LeafTerminal>(&terminal)) {
-        if (auto const x = std::get_if<IR::Term::LinkBlock>(e))
-            return fn(x->next);
-        if (auto const x = std::get_if<IR::Term::LinkBlockFast>(e))
-            return fn(x->next);
-    }
-    if (auto const x = std::get_if<IR::Term::If>(&terminal))
-        return AnyLocationDescriptorForTerminalHas(x->then_, fn) || AnyLocationDescriptorForTerminalHas(x->else_, fn);
-    if (auto const x = std::get_if<IR::Term::CheckBit>(&terminal))
-        return AnyLocationDescriptorForTerminalHas(x->then_, fn) || AnyLocationDescriptorForTerminalHas(x->else_, fn);
-    if (auto const x = std::get_if<IR::Term::CheckHalt>(&terminal))
-        return AnyLocationDescriptorForTerminalHas(x->else_, fn);
-    return false;
+bool AnyLocationDescriptorForTerminalHas(IR::Terminal terminal, Fn fn) {
+    return boost::apply_visitor([&](auto t) -> bool {
+        using T = std::decay_t<decltype(t)>;
+        if constexpr (std::is_same_v<T, IR::Term::Invalid>) {
+            return false;
+        } else if constexpr (std::is_same_v<T, IR::Term::ReturnToDispatch>) {
+            return false;
+        } else if constexpr (std::is_same_v<T, IR::Term::LinkBlock>) {
+            return fn(t.next);
+        } else if constexpr (std::is_same_v<T, IR::Term::LinkBlockFast>) {
+            return fn(t.next);
+        } else if constexpr (std::is_same_v<T, IR::Term::PopRSBHint>) {
+            return false;
+        } else if constexpr (std::is_same_v<T, IR::Term::FastDispatchHint>) {
+            return false;
+        } else if constexpr (std::is_same_v<T, IR::Term::If>) {
+            return AnyLocationDescriptorForTerminalHas(t.then_, fn) || AnyLocationDescriptorForTerminalHas(t.else_, fn);
+        } else if constexpr (std::is_same_v<T, IR::Term::CheckBit>) {
+            return AnyLocationDescriptorForTerminalHas(t.then_, fn) || AnyLocationDescriptorForTerminalHas(t.else_, fn);
+        } else if constexpr (std::is_same_v<T, IR::Term::CheckHalt>) {
+            return AnyLocationDescriptorForTerminalHas(t.else_, fn);
+        } else {
+            ASSERT(false && "Invalid terminal type");
+            return false;
+        }
+    }, terminal);
 }
 
 bool ShouldTestInst(u32 instruction, u32 pc, bool is_thumb, bool is_last_inst, A32::ITState it_state = {}) {

src/video_core/buffer_cache/buffer_cache.h

@@ -175,9 +175,71 @@ std::optional<VideoCore::RasterizerDownloadArea> BufferCache<P>::GetFlushArea(DA
 
 template <class P>
 void BufferCache<P>::DownloadMemory(DAddr device_addr, u64 size) {
-    ForEachBufferInRange(device_addr, size, [&](BufferId, Buffer& buffer) {
-        DownloadBufferMemory(buffer, device_addr, size);
+    if constexpr (!USE_MEMORY_MAPS) {
+        std::scoped_lock lock{mutex};
+        ForEachBufferInRange(device_addr, size, [&](BufferId, Buffer& buffer) {
+            DownloadBufferMemory(buffer, device_addr, size);
+        });
+        return;
+    }
+
+    boost::container::small_vector<std::pair<BufferCopy, BufferId>, 8> downloads;
+    u64 total_size_bytes = 0;
+    u64 largest_copy = 0;
+
+    std::unique_lock lock{mutex};
+    ForEachBufferInRange(device_addr, size, [&](BufferId buffer_id, Buffer& buffer) {
+        memory_tracker.ForEachDownloadRangeAndClear(
+            device_addr, size, [&](u64 device_addr_out, u64 range_size) {
+                const DAddr buffer_addr = buffer.CpuAddr();
+                const auto add_download = [&](DAddr start, DAddr end) {
+                    const u64 new_offset = start - buffer_addr;
+                    const u64 new_size = end - start;
+                    downloads.push_back({
+                        BufferCopy{
+                            .src_offset = new_offset,
+                            .dst_offset = total_size_bytes,
+                            .size = new_size,
+                        },
+                        buffer_id,
+                    });
+                    constexpr u64 align = 64ULL;
+                    constexpr u64 mask = ~(align - 1ULL);
+                    total_size_bytes += (new_size + align - 1) & mask;
+                    largest_copy = (std::max)(largest_copy, new_size);
+                };
+                gpu_modified_ranges.ForEachInRange(device_addr_out, range_size, add_download);
+                ClearDownload(device_addr_out, range_size);
+                gpu_modified_ranges.Subtract(device_addr_out, range_size);
+            });
     });
+    if (total_size_bytes == 0) {
+        return;
+    }
+
+    auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes);
+    boost::container::small_vector<BufferCopy, 8> writebacks;
+    runtime.PreCopyBarrier();
+    for (auto& [copy, buffer_id] : downloads) {
+        copy.dst_offset += download_staging.offset;
+        Buffer& buffer = slot_buffers[buffer_id];
+        buffer.MarkUsage(copy.src_offset, copy.size);
+        const std::array copies{copy};
+        runtime.CopyBuffer(download_staging.buffer, buffer, copies, false);
+        BufferCopy writeback{copy};
+        writeback.src_offset = static_cast<u64>(buffer.CpuAddr()) + copy.src_offset;
+        writebacks.push_back(writeback);
+    }
+    runtime.PostCopyBarrier();
+    lock.unlock();
+
+    runtime.Finish();
+    const u8* const base = download_staging.mapped_span.data();
+    for (const BufferCopy& writeback : writebacks) {
+        const u64 staging_offset = writeback.dst_offset - download_staging.offset;
+        device_memory.WriteBlockUnsafe(static_cast<DAddr>(writeback.src_offset),
+                                       base + staging_offset, writeback.size);
+    }
 }
 
 template <class P>
@@ -214,7 +276,7 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
     auto& src_buffer = slot_buffers[buffer_a];
     auto& dest_buffer = slot_buffers[buffer_b];
     SynchronizeBuffer(src_buffer, *cpu_src_address, static_cast<u32>(amount));
-    SynchronizeBuffer(dest_buffer, *cpu_dest_address, static_cast<u32>(amount));
+    memory_tracker.UnmarkRegionAsCpuModified(*cpu_dest_address, static_cast<u32>(amount));
     std::array copies{BufferCopy{
         .src_offset = src_buffer.Offset(*cpu_src_address),
         .dst_offset = dest_buffer.Offset(*cpu_dest_address),
@@ -673,32 +735,44 @@ void BufferCache<P>::PopAsyncFlushes() {
 
 template <class P>
 void BufferCache<P>::PopAsyncBuffers() {
-    if (async_buffers.empty()) {
-        return;
-    }
-    if (!async_buffers.front().has_value()) {
+    struct Writeback {
+        DAddr addr;
+        const u8* src;
+        u64 size;
+    };
+    boost::container::small_vector<Writeback, 8> writebacks;
+    {
+        std::scoped_lock lock{mutex};
+        if (async_buffers.empty()) {
+            return;
+        }
+        if (!async_buffers.front().has_value()) {
+            async_buffers.pop_front();
+            return;
+        }
+        auto& downloads = pending_downloads.front();
+        auto& async_buffer = async_buffers.front();
+        const u8* base = async_buffer->mapped_span.data();
+        const size_t base_offset = async_buffer->offset;
+        for (const auto& copy : downloads) {
+            const DAddr device_addr = static_cast<DAddr>(copy.src_offset);
+            const u64 dst_offset = copy.dst_offset - base_offset;
+            const u8* read_mapped_memory = base + dst_offset;
+            async_downloads.ForEachInRange(device_addr, copy.size, [&](DAddr start, DAddr end, s32) {
+                writebacks.push_back(
+                    {start, &read_mapped_memory[start - device_addr], end - start});
+            });
+            async_downloads.Subtract(device_addr, copy.size, [&](DAddr start, DAddr end) {
+                gpu_modified_ranges.Subtract(start, end - start);
+            });
+        }
+        async_buffers_death_ring.emplace_back(*async_buffer);
         async_buffers.pop_front();
-        return;
+        pending_downloads.pop_front();
     }
-    auto& downloads = pending_downloads.front();
-    auto& async_buffer = async_buffers.front();
-    u8* base = async_buffer->mapped_span.data();
-    const size_t base_offset = async_buffer->offset;
-    for (const auto& copy : downloads) {
-        const DAddr device_addr = static_cast<DAddr>(copy.src_offset);
-        const u64 dst_offset = copy.dst_offset - base_offset;
-        const u8* read_mapped_memory = base + dst_offset;
-        async_downloads.ForEachInRange(device_addr, copy.size, [&](DAddr start, DAddr end, s32) {
-            device_memory.WriteBlockUnsafe(start, &read_mapped_memory[start - device_addr],
-                                           end - start);
-        });
-        async_downloads.Subtract(device_addr, copy.size, [&](DAddr start, DAddr end) {
-            gpu_modified_ranges.Subtract(start, end - start);
-        });
+    for (const auto& wb : writebacks) {
+        device_memory.WriteBlockUnsafe(wb.addr, wb.src, wb.size);
     }
-    async_buffers_death_ring.emplace_back(*async_buffer);
-    async_buffers.pop_front();
-    pending_downloads.pop_front();
 }
 
 template <class P>

src/video_core/fence_manager.h

@@ -91,9 +91,6 @@ public:
             func();
         }
         fences.push(std::move(new_fence));
-        if (should_flush) {
-            rasterizer.FlushCommands();
-        }
         if constexpr (can_async_check) {
             guard.unlock();
             cv.notify_all();
@@ -238,10 +235,10 @@ private:
 
     void PopAsyncFlushes() {
         {
-            std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
+            std::scoped_lock lock{texture_cache.mutex};
             texture_cache.PopAsyncFlushes();
-            buffer_cache.PopAsyncFlushes();
         }
+        buffer_cache.PopAsyncFlushes();
         query_cache.PopAsyncFlushes();
     }
 

src/video_core/memory_manager.cpp

@@ -58,8 +58,9 @@ MemoryManager::MemoryManager(Core::System& system_, u64 address_space_bits_, GPU
 
 MemoryManager::~MemoryManager() = default;
 
-MemoryManager::EntryType MemoryManager::GetEntry(size_t position, bool is_big_page) const {
-    if (is_big_page) {
+template <bool is_big_page>
+MemoryManager::EntryType MemoryManager::GetEntry(size_t position) const {
+    if constexpr (is_big_page) {
         position = position >> big_page_bits;
         const u64 entry_mask = big_entries[position / 32];
         const size_t sub_index = position % 32;
@@ -72,8 +73,9 @@ MemoryManager::EntryType MemoryManager::GetEntry(size_t position, bool is_big_pa
     }
 }
 
-void MemoryManager::SetEntry(size_t position, MemoryManager::EntryType entry, bool is_big_page) {
-    if (is_big_page) {
+template <bool is_big_page>
+void MemoryManager::SetEntry(size_t position, MemoryManager::EntryType entry) {
+    if constexpr (is_big_page) {
         position = position >> big_page_bits;
         const u64 entry_mask = big_entries[position / 32];
         const size_t sub_index = position % 32;
@@ -106,21 +108,23 @@ inline void MemoryManager::SetBigPageContinuous(size_t big_page_index, bool valu
         (~(1ULL << sub_index) & continuous_mask) | (value ? 1ULL << sub_index : 0);
 }
 
-GPUVAddr MemoryManager::PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr, size_t size, PTEKind kind, MemoryManager::EntryType entry_type) {
+template <MemoryManager::EntryType entry_type>
+GPUVAddr MemoryManager::PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr, size_t size,
+                                    PTEKind kind) {
     [[maybe_unused]] u64 remaining_size{size};
-    if (entry_type == EntryType::Mapped) {
+    if constexpr (entry_type == EntryType::Mapped) {
         page_table.ReserveRange(gpu_addr, size);
     }
     for (u64 offset{}; offset < size; offset += page_size) {
         const GPUVAddr current_gpu_addr = gpu_addr + offset;
-        [[maybe_unused]] const auto current_entry_type = GetEntry(current_gpu_addr, false);
-        SetEntry(current_gpu_addr, entry_type, false);
+        [[maybe_unused]] const auto current_entry_type = GetEntry<false>(current_gpu_addr);
+        SetEntry<false>(current_gpu_addr, entry_type);
         if (current_entry_type != entry_type) {
             rasterizer->ModifyGPUMemory(unique_identifier, current_gpu_addr, page_size);
         }
-        if (entry_type == EntryType::Mapped) {
+        if constexpr (entry_type == EntryType::Mapped) {
             const DAddr current_dev_addr = dev_addr + offset;
-            const auto index = PageEntryIndex(current_gpu_addr, false);
+            const auto index = PageEntryIndex<false>(current_gpu_addr);
             const u32 sub_value = static_cast<u32>(current_dev_addr >> cpu_page_bits);
             page_table[index] = sub_value;
         }
@@ -130,18 +134,20 @@ GPUVAddr MemoryManager::PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr de
     return gpu_addr;
 }
 
-GPUVAddr MemoryManager::BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr, size_t size, PTEKind kind, MemoryManager::EntryType entry_type) {
+template <MemoryManager::EntryType entry_type>
+GPUVAddr MemoryManager::BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr,
+                                       size_t size, PTEKind kind) {
     [[maybe_unused]] u64 remaining_size{size};
     for (u64 offset{}; offset < size; offset += big_page_size) {
         const GPUVAddr current_gpu_addr = gpu_addr + offset;
-        [[maybe_unused]] const auto current_entry_type = GetEntry(current_gpu_addr, true);
-        SetEntry(current_gpu_addr, entry_type, true);
+        [[maybe_unused]] const auto current_entry_type = GetEntry<true>(current_gpu_addr);
+        SetEntry<true>(current_gpu_addr, entry_type);
         if (current_entry_type != entry_type) {
             rasterizer->ModifyGPUMemory(unique_identifier, current_gpu_addr, big_page_size);
         }
-        if (entry_type == EntryType::Mapped) {
+        if constexpr (entry_type == EntryType::Mapped) {
             const DAddr current_dev_addr = dev_addr + offset;
-            const auto index = PageEntryIndex(current_gpu_addr, true);
+            const auto index = PageEntryIndex<true>(current_gpu_addr);
             const u32 sub_value = static_cast<u32>(current_dev_addr >> cpu_page_bits);
             big_page_table_dev[index] = sub_value;
             const bool is_continuous = ([&] {
@@ -175,16 +181,19 @@ void MemoryManager::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_)
     rasterizer = rasterizer_;
 }
 
-GPUVAddr MemoryManager::Map(GPUVAddr gpu_addr, DAddr dev_addr, std::size_t size, PTEKind kind, bool is_big_pages) {
-    if (is_big_pages)
-        return BigPageTableOp(gpu_addr, dev_addr, size, kind, EntryType::Mapped);
-    return PageTableOp(gpu_addr, dev_addr, size, kind, EntryType::Mapped);
+GPUVAddr MemoryManager::Map(GPUVAddr gpu_addr, DAddr dev_addr, std::size_t size, PTEKind kind,
+                            bool is_big_pages) {
+    if (is_big_pages) [[likely]] {
+        return BigPageTableOp<EntryType::Mapped>(gpu_addr, dev_addr, size, kind);
+    }
+    return PageTableOp<EntryType::Mapped>(gpu_addr, dev_addr, size, kind);
 }
 
 GPUVAddr MemoryManager::MapSparse(GPUVAddr gpu_addr, std::size_t size, bool is_big_pages) {
-    if (is_big_pages)
-        return BigPageTableOp(gpu_addr, 0, size, PTEKind::INVALID, EntryType::Reserved);
-    return PageTableOp(gpu_addr, 0, size, PTEKind::INVALID, EntryType::Reserved);
+    if (is_big_pages) [[likely]] {
+        return BigPageTableOp<EntryType::Reserved>(gpu_addr, 0, size, PTEKind::INVALID);
+    }
+    return PageTableOp<EntryType::Reserved>(gpu_addr, 0, size, PTEKind::INVALID);
 }
 
 void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
@@ -198,21 +207,26 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
     }
     page_stash.clear();
 
-    BigPageTableOp(gpu_addr, 0, size, PTEKind::INVALID, EntryType::Free);
-    PageTableOp(gpu_addr, 0, size, PTEKind::INVALID, EntryType::Free);
+    BigPageTableOp<EntryType::Free>(gpu_addr, 0, size, PTEKind::INVALID);
+    PageTableOp<EntryType::Free>(gpu_addr, 0, size, PTEKind::INVALID);
 }
 
 std::optional<DAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) const {
     if (!IsWithinGPUAddressRange(gpu_addr)) [[unlikely]] {
         return std::nullopt;
     }
-    if (GetEntry(gpu_addr, true) != EntryType::Mapped) [[unlikely]] {
-        if (GetEntry(gpu_addr, false) != EntryType::Mapped)
+    if (GetEntry<true>(gpu_addr) != EntryType::Mapped) [[unlikely]] {
+        if (GetEntry<false>(gpu_addr) != EntryType::Mapped) {
             return std::nullopt;
-        const DAddr dev_addr_base = DAddr(page_table[PageEntryIndex(gpu_addr, false)]) << cpu_page_bits;
+        }
+
+        const DAddr dev_addr_base = static_cast<DAddr>(page_table[PageEntryIndex<false>(gpu_addr)])
+                                    << cpu_page_bits;
         return dev_addr_base + (gpu_addr & page_mask);
     }
-    const DAddr dev_addr_base = DAddr(big_page_table_dev[PageEntryIndex(gpu_addr, true)]) << cpu_page_bits;
+
+    const DAddr dev_addr_base =
+        static_cast<DAddr>(big_page_table_dev[PageEntryIndex<true>(gpu_addr)]) << cpu_page_bits;
     return dev_addr_base + (gpu_addr & big_page_mask);
 }
 
@@ -285,8 +299,10 @@ const u8* MemoryManager::GetPointer(GPUVAddr gpu_addr) const {
 #pragma inline_recursion(on)
 #endif
 
-template <typename FuncMapped, typename FuncReserved, typename FuncUnmapped>
-inline void MemoryManager::MemoryOperation(GPUVAddr gpu_src_addr, std::size_t size, bool is_big_page, FuncMapped&& func_mapped, FuncReserved&& func_reserved, FuncUnmapped&& func_unmapped) const {
+template <bool is_big_pages, typename FuncMapped, typename FuncReserved, typename FuncUnmapped>
+inline void MemoryManager::MemoryOperation(GPUVAddr gpu_src_addr, std::size_t size,
+                                           FuncMapped&& func_mapped, FuncReserved&& func_reserved,
+                                           FuncUnmapped&& func_unmapped) const {
     using FuncMappedReturn =
         typename std::invoke_result<FuncMapped, std::size_t, std::size_t, std::size_t>::type;
     using FuncReservedReturn =
@@ -299,7 +315,7 @@ inline void MemoryManager::MemoryOperation(GPUVAddr gpu_src_addr, std::size_t si
     u64 used_page_size;
     u64 used_page_mask;
     u64 used_page_bits;
-    if (is_big_page) {
+    if constexpr (is_big_pages) {
         used_page_size = big_page_size;
         used_page_mask = big_page_mask;
         used_page_bits = big_page_bits;
@@ -316,7 +332,7 @@ inline void MemoryManager::MemoryOperation(GPUVAddr gpu_src_addr, std::size_t si
     while (remaining_size > 0) {
         const std::size_t copy_amount{
             (std::min)(static_cast<std::size_t>(used_page_size) - page_offset, remaining_size)};
-        auto entry = GetEntry(current_address, is_big_page);
+        auto entry = GetEntry<is_big_pages>(current_address);
         if (entry == EntryType::Mapped) [[likely]] {
             if constexpr (BOOL_BREAK_MAPPED) {
                 if (func_mapped(page_index, page_offset, copy_amount)) {
@@ -351,14 +367,18 @@ inline void MemoryManager::MemoryOperation(GPUVAddr gpu_src_addr, std::size_t si
     }
 }
 
-void MemoryManager::ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size, [[maybe_unused]] VideoCommon::CacheType which, bool unsafe) const {
-    auto set_to_zero = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset, std::size_t copy_amount) {
+template <bool is_safe>
+void MemoryManager::ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size,
+                                  [[maybe_unused]] VideoCommon::CacheType which) const {
+    auto set_to_zero = [&]([[maybe_unused]] std::size_t page_index,
+                           [[maybe_unused]] std::size_t offset, std::size_t copy_amount) {
         std::memset(dest_buffer, 0, copy_amount);
         dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount;
     };
     auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
-        const DAddr dev_addr_base = (DAddr(page_table[page_index]) << cpu_page_bits) + offset;
-        if (!unsafe) {
+        const DAddr dev_addr_base =
+            (static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
+        if constexpr (is_safe) {
             rasterizer->FlushRegion(dev_addr_base, copy_amount, which);
         }
         u8* physical = memory.GetPointer<u8>(dev_addr_base);
@@ -366,8 +386,9 @@ void MemoryManager::ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std:
         dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount;
     };
     auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
-        const DAddr dev_addr_base = (DAddr(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
-        if (!unsafe) {
+        const DAddr dev_addr_base =
+            (static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
+        if constexpr (is_safe) {
             rasterizer->FlushRegion(dev_addr_base, copy_amount, which);
         }
         if (!IsBigPageContinuous(page_index)) [[unlikely]] {
@@ -378,28 +399,35 @@ void MemoryManager::ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std:
         }
         dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount;
     };
-    auto read_short_pages = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
+    auto read_short_pages = [&](std::size_t page_index, std::size_t offset,
+                                std::size_t copy_amount) {
         GPUVAddr base = (page_index << big_page_bits) + offset;
-        MemoryOperation(base, copy_amount, false, mapped_normal, set_to_zero, set_to_zero);
+        MemoryOperation<false>(base, copy_amount, mapped_normal, set_to_zero, set_to_zero);
     };
-    MemoryOperation(gpu_src_addr, size, true, mapped_big, set_to_zero, read_short_pages);
+    MemoryOperation<true>(gpu_src_addr, size, mapped_big, set_to_zero, read_short_pages);
 }
 
-void MemoryManager::ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size, VideoCommon::CacheType which) const {
-    ReadBlockImpl(gpu_src_addr, dest_buffer, size, which, false);
+void MemoryManager::ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size,
+                              VideoCommon::CacheType which) const {
+    ReadBlockImpl<true>(gpu_src_addr, dest_buffer, size, which);
 }
 
-void MemoryManager::ReadBlockUnsafe(GPUVAddr gpu_src_addr, void* dest_buffer, const std::size_t size) const {
-    ReadBlockImpl(gpu_src_addr, dest_buffer, size, VideoCommon::CacheType::None, true);
+void MemoryManager::ReadBlockUnsafe(GPUVAddr gpu_src_addr, void* dest_buffer,
+                                    const std::size_t size) const {
+    ReadBlockImpl<false>(gpu_src_addr, dest_buffer, size, VideoCommon::CacheType::None);
 }
 
-void MemoryManager::WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size, [[maybe_unused]] VideoCommon::CacheType which, bool unsafe) {
-    auto just_advance = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset, std::size_t copy_amount) {
+template <bool is_safe>
+void MemoryManager::WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size,
+                                   [[maybe_unused]] VideoCommon::CacheType which) {
+    auto just_advance = [&]([[maybe_unused]] std::size_t page_index,
+                            [[maybe_unused]] std::size_t offset, std::size_t copy_amount) {
         src_buffer = static_cast<const u8*>(src_buffer) + copy_amount;
     };
     auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
-        const DAddr dev_addr_base = (DAddr(page_table[page_index]) << cpu_page_bits) + offset;
-        if (!unsafe) {
+        const DAddr dev_addr_base =
+            (static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
+        if constexpr (is_safe) {
             rasterizer->InvalidateRegion(dev_addr_base, copy_amount, which);
         }
         u8* physical = memory.GetPointer<u8>(dev_addr_base);
@@ -407,8 +435,9 @@ void MemoryManager::WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffe
         src_buffer = static_cast<const u8*>(src_buffer) + copy_amount;
     };
     auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
-        const DAddr dev_addr_base = (DAddr(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
-        if (!unsafe) {
+        const DAddr dev_addr_base =
+            (static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
+        if constexpr (is_safe) {
             rasterizer->InvalidateRegion(dev_addr_base, copy_amount, which);
         }
         if (!IsBigPageContinuous(page_index)) [[unlikely]] {
@@ -419,23 +448,26 @@ void MemoryManager::WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffe
         }
         src_buffer = static_cast<const u8*>(src_buffer) + copy_amount;
     };
-    auto write_short_pages = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
+    auto write_short_pages = [&](std::size_t page_index, std::size_t offset,
+                                 std::size_t copy_amount) {
         GPUVAddr base = (page_index << big_page_bits) + offset;
-        MemoryOperation(base, copy_amount, false, mapped_normal, just_advance, just_advance);
+        MemoryOperation<false>(base, copy_amount, mapped_normal, just_advance, just_advance);
     };
-    MemoryOperation(gpu_dest_addr, size, true, mapped_big, just_advance, write_short_pages);
+    MemoryOperation<true>(gpu_dest_addr, size, mapped_big, just_advance, write_short_pages);
 }
 
-void MemoryManager::WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size, VideoCommon::CacheType which) {
-    WriteBlockImpl(gpu_dest_addr, src_buffer, size, which, false);
+void MemoryManager::WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size,
+                               VideoCommon::CacheType which) {
+    WriteBlockImpl<true>(gpu_dest_addr, src_buffer, size, which);
 }
 
-void MemoryManager::WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size) {
-    WriteBlockImpl(gpu_dest_addr, src_buffer, size, VideoCommon::CacheType::None, true);
+void MemoryManager::WriteBlockUnsafe(GPUVAddr gpu_dest_addr, const void* src_buffer,
+                                     std::size_t size) {
+    WriteBlockImpl<false>(gpu_dest_addr, src_buffer, size, VideoCommon::CacheType::None);
 }
 
 void MemoryManager::WriteBlockCached(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size) {
-    WriteBlockImpl(gpu_dest_addr, src_buffer, size, VideoCommon::CacheType::None, true);
+    WriteBlockImpl<false>(gpu_dest_addr, src_buffer, size, VideoCommon::CacheType::None);
     accumulator.Add(gpu_dest_addr, size);
 }
 
@@ -446,18 +478,21 @@ void MemoryManager::FlushRegion(GPUVAddr gpu_addr, size_t size,
                           [[maybe_unused]] std::size_t copy_amount) {};
 
     auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
-        const DAddr dev_addr_base = (DAddr(page_table[page_index]) << cpu_page_bits) + offset;
+        const DAddr dev_addr_base =
+            (static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
         rasterizer->FlushRegion(dev_addr_base, copy_amount, which);
     };
     auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
-        const DAddr dev_addr_base = (DAddr(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
+        const DAddr dev_addr_base =
+            (static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
         rasterizer->FlushRegion(dev_addr_base, copy_amount, which);
     };
-    auto flush_short_pages = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
+    auto flush_short_pages = [&](std::size_t page_index, std::size_t offset,
+                                 std::size_t copy_amount) {
         GPUVAddr base = (page_index << big_page_bits) + offset;
-        MemoryOperation(base, copy_amount, false, mapped_normal, do_nothing, do_nothing);
+        MemoryOperation<false>(base, copy_amount, mapped_normal, do_nothing, do_nothing);
     };
-    MemoryOperation(gpu_addr, size, true, mapped_big, do_nothing, flush_short_pages);
+    MemoryOperation<true>(gpu_addr, size, mapped_big, do_nothing, flush_short_pages);
 }
 
 bool MemoryManager::IsMemoryDirty(GPUVAddr gpu_addr, size_t size,
@@ -482,10 +517,10 @@ bool MemoryManager::IsMemoryDirty(GPUVAddr gpu_addr, size_t size,
     auto check_short_pages = [&](std::size_t page_index, std::size_t offset,
                                  std::size_t copy_amount) {
         GPUVAddr base = (page_index << big_page_bits) + offset;
-        MemoryOperation(base, copy_amount, false, mapped_normal, do_nothing, do_nothing);
+        MemoryOperation<false>(base, copy_amount, mapped_normal, do_nothing, do_nothing);
         return result;
     };
-    MemoryOperation(gpu_addr, size, true, mapped_big, do_nothing, check_short_pages);
+    MemoryOperation<true>(gpu_addr, size, mapped_big, do_nothing, check_short_pages);
     return result;
 }
 
@@ -522,10 +557,10 @@ size_t MemoryManager::MaxContinuousRange(GPUVAddr gpu_addr, size_t size) const {
     auto check_short_pages = [&](std::size_t page_index, std::size_t offset,
                                  std::size_t copy_amount) {
         GPUVAddr base = (page_index << big_page_bits) + offset;
-        MemoryOperation(base, copy_amount, false, short_check, fail, fail);
+        MemoryOperation<false>(base, copy_amount, short_check, fail, fail);
         return result;
     };
-    MemoryOperation(gpu_addr, size, true, big_check, fail, check_short_pages);
+    MemoryOperation<true>(gpu_addr, size, big_check, fail, check_short_pages);
     return range_so_far;
 }
 
@@ -541,18 +576,21 @@ void MemoryManager::InvalidateRegion(GPUVAddr gpu_addr, size_t size,
                           [[maybe_unused]] std::size_t copy_amount) {};
 
     auto mapped_normal = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
-        const DAddr dev_addr_base = (DAddr(page_table[page_index]) << cpu_page_bits) + offset;
+        const DAddr dev_addr_base =
+            (static_cast<DAddr>(page_table[page_index]) << cpu_page_bits) + offset;
         rasterizer->InvalidateRegion(dev_addr_base, copy_amount, which);
     };
     auto mapped_big = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
-        const DAddr dev_addr_base = (DAddr(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
+        const DAddr dev_addr_base =
+            (static_cast<DAddr>(big_page_table_dev[page_index]) << cpu_page_bits) + offset;
         rasterizer->InvalidateRegion(dev_addr_base, copy_amount, which);
     };
-    auto invalidate_short_pages = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
+    auto invalidate_short_pages = [&](std::size_t page_index, std::size_t offset,
+                                      std::size_t copy_amount) {
         GPUVAddr base = (page_index << big_page_bits) + offset;
-        MemoryOperation(base, copy_amount, false, mapped_normal, do_nothing, do_nothing);
+        MemoryOperation<false>(base, copy_amount, mapped_normal, do_nothing, do_nothing);
     };
-    MemoryOperation(gpu_addr, size, true, mapped_big, do_nothing, invalidate_short_pages);
+    MemoryOperation<true>(gpu_addr, size, mapped_big, do_nothing, invalidate_short_pages);
 }
 
 void MemoryManager::CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std::size_t size,
@@ -564,7 +602,7 @@ void MemoryManager::CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std
 }
 
 bool MemoryManager::IsGranularRange(GPUVAddr gpu_addr, std::size_t size) const {
-    if (GetEntry(gpu_addr, true) == EntryType::Mapped) [[likely]] {
+    if (GetEntry<true>(gpu_addr) == EntryType::Mapped) [[likely]] {
         size_t page_index = gpu_addr >> big_page_bits;
         if (IsBigPageContinuous(page_index)) [[likely]] {
             const std::size_t page{(page_index & big_page_mask) + size};
@@ -573,7 +611,7 @@ bool MemoryManager::IsGranularRange(GPUVAddr gpu_addr, std::size_t size) const {
         const std::size_t page{(gpu_addr & Core::DEVICE_PAGEMASK) + size};
         return page <= Core::DEVICE_PAGESIZE;
     }
-    if (GetEntry(gpu_addr, false) != EntryType::Mapped) {
+    if (GetEntry<false>(gpu_addr) != EntryType::Mapped) {
         return false;
     }
     const std::size_t page{(gpu_addr & Core::DEVICE_PAGEMASK) + size};
@@ -611,10 +649,10 @@ bool MemoryManager::IsContinuousRange(GPUVAddr gpu_addr, std::size_t size) const
     auto check_short_pages = [&](std::size_t page_index, std::size_t offset,
                                  std::size_t copy_amount) {
         GPUVAddr base = (page_index << big_page_bits) + offset;
-        MemoryOperation(base, copy_amount, false, short_check, fail, fail);
+        MemoryOperation<false>(base, copy_amount, short_check, fail, fail);
         return !result;
     };
-    MemoryOperation(gpu_addr, size, true, big_check, fail, check_short_pages);
+    MemoryOperation<true>(gpu_addr, size, big_check, fail, check_short_pages);
     return result;
 }
 
@@ -627,12 +665,13 @@ bool MemoryManager::IsFullyMappedRange(GPUVAddr gpu_addr, std::size_t size) cons
     };
     auto pass = [&]([[maybe_unused]] std::size_t page_index, [[maybe_unused]] std::size_t offset,
                     [[maybe_unused]] std::size_t copy_amount) { return false; };
-    auto check_short_pages = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
+    auto check_short_pages = [&](std::size_t page_index, std::size_t offset,
+                                 std::size_t copy_amount) {
         GPUVAddr base = (page_index << big_page_bits) + offset;
-        MemoryOperation(base, copy_amount, false, pass, pass, fail);
+        MemoryOperation<false>(base, copy_amount, pass, pass, fail);
         return !result;
     };
-    MemoryOperation(gpu_addr, size, true, pass, fail, check_short_pages);
+    MemoryOperation<true>(gpu_addr, size, pass, fail, check_short_pages);
     return result;
 }
 
@@ -644,9 +683,13 @@ MemoryManager::GetSubmappedRange(GPUVAddr gpu_addr, std::size_t size) const {
 }
 
 template <bool is_gpu_address>
-void MemoryManager::GetSubmappedRangeImpl(GPUVAddr gpu_addr, std::size_t size, boost::container::small_vector<std::pair<std::conditional_t<is_gpu_address, GPUVAddr, DAddr>, std::size_t>, 32>& result)
+void MemoryManager::GetSubmappedRangeImpl(
+    GPUVAddr gpu_addr, std::size_t size,
+    boost::container::small_vector<
+        std::pair<std::conditional_t<is_gpu_address, GPUVAddr, DAddr>, std::size_t>, 32>& result)
     const {
-    std::optional<std::pair<std::conditional_t<is_gpu_address, GPUVAddr, DAddr>, std::size_t>> last_segment{};
+    std::optional<std::pair<std::conditional_t<is_gpu_address, GPUVAddr, DAddr>, std::size_t>>
+        last_segment{};
     std::optional<DAddr> old_page_addr{};
     const auto split = [&last_segment, &result]([[maybe_unused]] std::size_t page_index,
                                                 [[maybe_unused]] std::size_t offset,
@@ -702,9 +745,9 @@ void MemoryManager::GetSubmappedRangeImpl(GPUVAddr gpu_addr, std::size_t size, b
     };
     auto do_short_pages = [&](std::size_t page_index, std::size_t offset, std::size_t copy_amount) {
         GPUVAddr base = (page_index << big_page_bits) + offset;
-        MemoryOperation(base, copy_amount, false, extend_size_short, split, split);
+        MemoryOperation<false>(base, copy_amount, extend_size_short, split, split);
     };
-    MemoryOperation(gpu_addr, size, true, extend_size_big, split, do_short_pages);
+    MemoryOperation<true>(gpu_addr, size, extend_size_big, split, do_short_pages);
     split(0, 0, 0);
 }
 

src/video_core/memory_manager.h

@@ -45,7 +45,7 @@ public:
 
     static constexpr bool HAS_FLUSH_INVALIDATION = true;
 
-    inline size_t GetID() const noexcept {
+    size_t GetID() const {
         return unique_identifier;
     }
 
@@ -66,15 +66,16 @@ public:
     [[nodiscard]] const u8* GetPointer(GPUVAddr addr) const;
 
     template <typename T>
-    [[nodiscard]] inline T* GetPointer(GPUVAddr addr) noexcept {
-        const auto address = GpuToCpuAddress(addr);
-        if (!address)
+    [[nodiscard]] T* GetPointer(GPUVAddr addr) {
+        const auto address{GpuToCpuAddress(addr)};
+        if (!address) {
             return {};
+        }
         return memory.GetPointer<T>(*address);
     }
 
     template <typename T>
-    [[nodiscard]] inline const T* GetPointer(GPUVAddr addr) const noexcept {
+    [[nodiscard]] const T* GetPointer(GPUVAddr addr) const {
         return GetPointer<T*>(addr);
     }
 
@@ -84,9 +85,12 @@ public:
      * in the Host Memory counterpart. Note: This functions cause Host GPU Memory
      * Flushes and Invalidations, respectively to each operation.
      */
-    void ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size, VideoCommon::CacheType which = VideoCommon::CacheType::All) const;
-    void WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size, VideoCommon::CacheType which = VideoCommon::CacheType::All);
-    void CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std::size_t size, VideoCommon::CacheType which = VideoCommon::CacheType::All);
+    void ReadBlock(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size,
+                   VideoCommon::CacheType which = VideoCommon::CacheType::All) const;
+    void WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size,
+                    VideoCommon::CacheType which = VideoCommon::CacheType::All);
+    void CopyBlock(GPUVAddr gpu_dest_addr, GPUVAddr gpu_src_addr, std::size_t size,
+                   VideoCommon::CacheType which = VideoCommon::CacheType::All);
 
     /**
      * ReadBlockUnsafe and WriteBlockUnsafe are special versions of ReadBlock and
@@ -156,14 +160,21 @@ public:
     u8* GetSpan(const GPUVAddr src_addr, const std::size_t size);
 
 private:
-    template <typename FuncMapped, typename FuncReserved, typename FuncUnmapped>
-    inline void MemoryOperation(GPUVAddr gpu_src_addr, std::size_t size, bool is_big_page, FuncMapped&& func_mapped, FuncReserved&& func_reserved, FuncUnmapped&& func_unmapped) const;
+    template <bool is_big_pages, typename FuncMapped, typename FuncReserved, typename FuncUnmapped>
+    inline void MemoryOperation(GPUVAddr gpu_src_addr, std::size_t size, FuncMapped&& func_mapped,
+                                FuncReserved&& func_reserved, FuncUnmapped&& func_unmapped) const;
 
-    void ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size, VideoCommon::CacheType which, bool unsafe) const;
-    void WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size, VideoCommon::CacheType which, bool unsafe);
+    template <bool is_safe>
+    void ReadBlockImpl(GPUVAddr gpu_src_addr, void* dest_buffer, std::size_t size,
+                       VideoCommon::CacheType which) const;
 
-    [[nodiscard]] std::size_t PageEntryIndex(GPUVAddr gpu_addr, bool is_big_page) const {
-        if (is_big_page) {
+    template <bool is_safe>
+    void WriteBlockImpl(GPUVAddr gpu_dest_addr, const void* src_buffer, std::size_t size,
+                        VideoCommon::CacheType which);
+
+    template <bool is_big_page>
+    [[nodiscard]] std::size_t PageEntryIndex(GPUVAddr gpu_addr) const {
+        if constexpr (is_big_page) {
             return (gpu_addr >> big_page_bits) & big_page_table_mask;
         } else {
             return (gpu_addr >> page_bits) & page_table_mask;
@@ -176,7 +187,9 @@ private:
     template <bool is_gpu_address>
     void GetSubmappedRangeImpl(
         GPUVAddr gpu_addr, std::size_t size,
-        boost::container::small_vector<std::pair<std::conditional_t<is_gpu_address, GPUVAddr, DAddr>, std::size_t>, 32>& result) const;
+        boost::container::small_vector<
+            std::pair<std::conditional_t<is_gpu_address, GPUVAddr, DAddr>, std::size_t>, 32>&
+            result) const;
 
     Core::System& system;
     MaxwellDeviceMemoryManager& memory;
@@ -206,11 +219,19 @@ private:
     std::vector<u64> entries;
     std::vector<u64> big_entries;
 
-    GPUVAddr PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr, size_t size, PTEKind kind, EntryType entry_type);
-    GPUVAddr BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr, size_t size, PTEKind kind, EntryType entry_type);
+    template <EntryType entry_type>
+    GPUVAddr PageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr, size_t size,
+                         PTEKind kind);
 
-    inline EntryType GetEntry(size_t position, bool is_big_page) const;
-    inline void SetEntry(size_t position, EntryType entry, bool is_big_page);
+    template <EntryType entry_type>
+    GPUVAddr BigPageTableOp(GPUVAddr gpu_addr, [[maybe_unused]] DAddr dev_addr, size_t size,
+                            PTEKind kind);
+
+    template <bool is_big_page>
+    inline EntryType GetEntry(size_t position) const;
+
+    template <bool is_big_page>
+    inline void SetEntry(size_t position, EntryType entry);
 
     Common::MultiLevelPageTable<u32> page_table;
     Common::RangeMap<GPUVAddr, PTEKind> kind_map;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -485,7 +485,6 @@ void RasterizerOpenGL::FlushRegion(DAddr addr, u64 size, VideoCommon::CacheType
         texture_cache.DownloadMemory(addr, size);
     }
     if ((True(which & VideoCommon::CacheType::BufferCache))) {
-        std::scoped_lock lock{buffer_cache.mutex};
         buffer_cache.DownloadMemory(addr, size);
     }
     if ((True(which & VideoCommon::CacheType::QueryCache))) {

src/video_core/renderer_vulkan/blit_image.cpp

@@ -545,6 +545,10 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view)
 
 void BeginRenderPass(vk::CommandBuffer& cmdbuf, const Framebuffer* framebuffer) {
     const VkRenderPass render_pass = framebuffer->RenderPass();
+    if (!render_pass) {
+        framebuffer->BeginRendering(cmdbuf);
+        return;
+    }
     const VkFramebuffer framebuffer_handle = framebuffer->Handle();
     const VkExtent2D render_area = framebuffer->RenderArea();
     const VkRenderPassBeginInfo renderpass_bi{
@@ -561,6 +565,31 @@ void BeginRenderPass(vk::CommandBuffer& cmdbuf, const Framebuffer* framebuffer)
     };
     cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
 }
+
+void EndRenderPass(vk::CommandBuffer& cmdbuf, const Framebuffer* framebuffer) {
+    if (framebuffer->RenderPass()) {
+        cmdbuf.EndRenderPass();
+    } else {
+        cmdbuf.EndRendering();
+    }
+}
+
+[[nodiscard]] VkPipelineRenderingCreateInfo MakePipelineRenderingCreateInfo(
+    const Framebuffer* framebuffer) {
+    return VkPipelineRenderingCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_RENDERING_CREATE_INFO,
+        .pNext = nullptr,
+        .viewMask = 0,
+        .colorAttachmentCount = framebuffer->NumColorAttachments(),
+        .pColorAttachmentFormats = framebuffer->ColorAttachmentFormats().data(),
+        .depthAttachmentFormat = framebuffer->HasAspectDepthBit()
+                                     ? framebuffer->DepthAttachmentFormat()
+                                     : VK_FORMAT_UNDEFINED,
+        .stencilAttachmentFormat = framebuffer->HasAspectStencilBit()
+                                       ? framebuffer->DepthAttachmentFormat()
+                                       : VK_FORMAT_UNDEFINED,
+    };
+}
 } // Anonymous namespace
 
 BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_,
@@ -623,10 +652,12 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, const ImageV
     const BlitImagePipelineKey key{
         .renderpass = dst_framebuffer->RenderPass(),
         .operation = operation,
+        .color_formats = dst_framebuffer->ColorAttachmentFormats(),
+        .depth_format = dst_framebuffer->DepthAttachmentFormat(),
     };
     const VkPipelineLayout layout = *one_texture_pipeline_layout;
     const VkSampler sampler = is_linear ? *linear_sampler : *nearest_sampler;
-    const VkPipeline pipeline = FindOrEmplaceColorPipeline(key);
+    const VkPipeline pipeline = FindOrEmplaceColorPipeline(key, dst_framebuffer);
     const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D);
 
     RecordShaderReadBarrier(scheduler, src_image_view);
@@ -651,9 +682,11 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView
     const BlitImagePipelineKey key{
         .renderpass = dst_framebuffer->RenderPass(),
         .operation = Tegra::Engines::Fermi2D::Operation::SrcCopy,
+        .color_formats = dst_framebuffer->ColorAttachmentFormats(),
+        .depth_format = dst_framebuffer->DepthAttachmentFormat(),
     };
     const VkPipelineLayout layout = *one_texture_pipeline_layout;
-    const VkPipeline pipeline = FindOrEmplaceColorPipeline(key);
+    const VkPipeline pipeline = FindOrEmplaceColorPipeline(key, dst_framebuffer);
     scheduler.RequestOutsideRenderPassOperationContext();
     scheduler.Record([this, dst_framebuffer, src_image_view, src_image, src_sampler, dst_region,
                       src_region, src_size, pipeline, layout](vk::CommandBuffer cmdbuf) {
@@ -666,7 +699,7 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView
                                   nullptr);
         BindBlitState(cmdbuf, layout, dst_region, src_region, src_size);
         cmdbuf.Draw(3, 1, 0, 0);
-        cmdbuf.EndRenderPass();
+        EndRenderPass(cmdbuf, dst_framebuffer);
     });
 }
 
@@ -676,10 +709,12 @@ void BlitImageHelper::BlitColorMSAA(const Framebuffer* dst_framebuffer,
     const BlitMSAAPipelineKey key{
         .renderpass = dst_framebuffer->RenderPass(),
         .samples = dst_framebuffer->Samples(),
+        .color_formats = dst_framebuffer->ColorAttachmentFormats(),
+        .depth_format = dst_framebuffer->DepthAttachmentFormat(),
     };
     const VkPipelineLayout layout = *one_texture_pipeline_layout;
     const VkSampler sampler = *nearest_sampler;
-    const VkPipeline pipeline = FindOrEmplaceBlitColorMSAAPipeline(key);
+    const VkPipeline pipeline = FindOrEmplaceBlitColorMSAAPipeline(key, dst_framebuffer);
     const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D);
 
     RecordShaderReadBarrier(scheduler, src_image_view);
@@ -703,7 +738,7 @@ void BlitImageHelper::ResolveDepthStencil(const Framebuffer* dst_framebuffer,
     const bool resolve_stencil =
         dst_framebuffer->HasAspectStencilBit() && device.IsExtShaderStencilExportSupported();
     const VkPipeline pipeline =
-        FindOrEmplaceResolveDepthStencilPipeline(dst_framebuffer->RenderPass(), resolve_stencil);
+        FindOrEmplaceResolveDepthStencilPipeline(dst_framebuffer, resolve_stencil);
     const VkPipelineLayout layout =
         resolve_stencil ? *two_textures_pipeline_layout : *one_texture_pipeline_layout;
     const VkSampler sampler = *nearest_sampler;
@@ -747,10 +782,12 @@ void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer,
     const BlitImagePipelineKey key{
         .renderpass = dst_framebuffer->RenderPass(),
         .operation = operation,
+        .color_formats = dst_framebuffer->ColorAttachmentFormats(),
+        .depth_format = dst_framebuffer->DepthAttachmentFormat(),
     };
     const VkPipelineLayout layout = *two_textures_pipeline_layout;
     const VkSampler sampler = *nearest_sampler;
-    const VkPipeline pipeline = FindOrEmplaceDepthStencilPipeline(key);
+    const VkPipeline pipeline = FindOrEmplaceDepthStencilPipeline(key, dst_framebuffer);
     const VkImageView src_depth_view = src_image_view.DepthView();
     const VkImageView src_stencil_view = src_image_view.StencilView();
 
@@ -772,25 +809,25 @@ void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer,
 
 void BlitImageHelper::ConvertD32ToR32(const Framebuffer* dst_framebuffer,
                                       const ImageView& src_image_view) {
-    ConvertDepthToColorPipeline(convert_d32_to_r32_pipeline, dst_framebuffer->RenderPass());
+    ConvertDepthToColorPipeline(convert_d32_to_r32_pipeline, dst_framebuffer);
     Convert(*convert_d32_to_r32_pipeline, dst_framebuffer, src_image_view);
 }
 
 void BlitImageHelper::ConvertR32ToD32(const Framebuffer* dst_framebuffer,
                                       const ImageView& src_image_view) {
-    ConvertColorToDepthPipeline(convert_r32_to_d32_pipeline, dst_framebuffer->RenderPass());
+    ConvertColorToDepthPipeline(convert_r32_to_d32_pipeline, dst_framebuffer);
     Convert(*convert_r32_to_d32_pipeline, dst_framebuffer, src_image_view);
 }
 
 void BlitImageHelper::ConvertD16ToR16(const Framebuffer* dst_framebuffer,
                                       const ImageView& src_image_view) {
-    ConvertDepthToColorPipeline(convert_d16_to_r16_pipeline, dst_framebuffer->RenderPass());
+    ConvertDepthToColorPipeline(convert_d16_to_r16_pipeline, dst_framebuffer);
     Convert(*convert_d16_to_r16_pipeline, dst_framebuffer, src_image_view);
 }
 
 void BlitImageHelper::ConvertR16ToD16(const Framebuffer* dst_framebuffer,
                                       const ImageView& src_image_view) {
-    ConvertColorToDepthPipeline(convert_r16_to_d16_pipeline, dst_framebuffer->RenderPass());
+    ConvertColorToDepthPipeline(convert_r16_to_d16_pipeline, dst_framebuffer);
     Convert(*convert_r16_to_d16_pipeline, dst_framebuffer, src_image_view);
 }
 
@@ -801,35 +838,35 @@ void BlitImageHelper::ConvertABGR8ToD24S8(const Framebuffer* dst_framebuffer,
         LOG_WARNING(Render_Vulkan, "ConvertABGR8ToD24S8 requires shader_stencil_export, skipping");
         return;
     }
-    ConvertPipelineDepthTargetEx(convert_abgr8_to_d24s8_pipeline, dst_framebuffer->RenderPass(),
+    ConvertPipelineDepthTargetEx(convert_abgr8_to_d24s8_pipeline, dst_framebuffer,
                                  convert_abgr8_to_d24s8_frag);
     Convert(*convert_abgr8_to_d24s8_pipeline, dst_framebuffer, src_image_view);
 }
 
 void BlitImageHelper::ConvertABGR8ToD32F(const Framebuffer* dst_framebuffer,
                                          const ImageView& src_image_view) {
-    ConvertPipelineDepthTargetEx(convert_abgr8_to_d32f_pipeline, dst_framebuffer->RenderPass(),
+    ConvertPipelineDepthTargetEx(convert_abgr8_to_d32f_pipeline, dst_framebuffer,
                                  convert_abgr8_to_d32f_frag);
     Convert(*convert_abgr8_to_d32f_pipeline, dst_framebuffer, src_image_view);
 }
 
 void BlitImageHelper::ConvertD32FToABGR8(const Framebuffer* dst_framebuffer,
                                          ImageView& src_image_view) {
-    ConvertPipelineColorTargetEx(convert_d32f_to_abgr8_pipeline, dst_framebuffer->RenderPass(),
+    ConvertPipelineColorTargetEx(convert_d32f_to_abgr8_pipeline, dst_framebuffer,
                                  convert_d32f_to_abgr8_frag);
     ConvertDepthStencil(*convert_d32f_to_abgr8_pipeline, dst_framebuffer, src_image_view);
 }
 
 void BlitImageHelper::ConvertD24S8ToABGR8(const Framebuffer* dst_framebuffer,
                                           ImageView& src_image_view) {
-    ConvertPipelineColorTargetEx(convert_d24s8_to_abgr8_pipeline, dst_framebuffer->RenderPass(),
+    ConvertPipelineColorTargetEx(convert_d24s8_to_abgr8_pipeline, dst_framebuffer,
                                  convert_d24s8_to_abgr8_frag);
     ConvertDepthStencil(*convert_d24s8_to_abgr8_pipeline, dst_framebuffer, src_image_view);
 }
 
 void BlitImageHelper::ConvertS8D24ToABGR8(const Framebuffer* dst_framebuffer,
                                           ImageView& src_image_view) {
-    ConvertPipelineColorTargetEx(convert_s8d24_to_abgr8_pipeline, dst_framebuffer->RenderPass(),
+    ConvertPipelineColorTargetEx(convert_s8d24_to_abgr8_pipeline, dst_framebuffer,
                                  convert_s8d24_to_abgr8_frag);
     ConvertDepthStencil(*convert_s8d24_to_abgr8_pipeline, dst_framebuffer, src_image_view);
 }
@@ -840,8 +877,10 @@ void BlitImageHelper::ClearColor(const Framebuffer* dst_framebuffer, u8 color_ma
     const BlitImagePipelineKey key{
         .renderpass = dst_framebuffer->RenderPass(),
         .operation = Tegra::Engines::Fermi2D::Operation::BlendPremult,
+        .color_formats = dst_framebuffer->ColorAttachmentFormats(),
+        .depth_format = dst_framebuffer->DepthAttachmentFormat(),
     };
-    const VkPipeline pipeline = FindOrEmplaceClearColorPipeline(key);
+    const VkPipeline pipeline = FindOrEmplaceClearColorPipeline(key, dst_framebuffer);
     const VkPipelineLayout layout = *clear_color_pipeline_layout;
     scheduler.RequestRenderpass(dst_framebuffer);
     scheduler.Record(
@@ -867,8 +906,10 @@ void BlitImageHelper::ClearDepthStencil(const Framebuffer* dst_framebuffer, bool
         .stencil_mask = stencil_mask,
         .stencil_compare_mask = stencil_compare_mask,
         .stencil_ref = stencil_ref,
+        .color_formats = dst_framebuffer->ColorAttachmentFormats(),
+        .depth_format = dst_framebuffer->DepthAttachmentFormat(),
     };
-    const VkPipeline pipeline = FindOrEmplaceClearStencilPipeline(key);
+    const VkPipeline pipeline = FindOrEmplaceClearStencilPipeline(key, dst_framebuffer);
     const VkPipelineLayout layout = *clear_color_pipeline_layout;
     scheduler.RequestRenderpass(dst_framebuffer);
     scheduler.Record([pipeline, layout, clear_depth, dst_region](vk::CommandBuffer cmdbuf) {
@@ -1140,12 +1181,14 @@ void BlitImageHelper::ConvertDepthStencil(VkPipeline pipeline, const Framebuffer
     scheduler.InvalidateState();
 }
 
-VkPipeline BlitImageHelper::FindOrEmplaceColorPipeline(const BlitImagePipelineKey& key) {
+VkPipeline BlitImageHelper::FindOrEmplaceColorPipeline(const BlitImagePipelineKey& key,
+                                                      const Framebuffer* framebuffer) {
     const auto it = std::ranges::find(blit_color_keys, key);
     if (it != blit_color_keys.end()) {
         return *blit_color_pipelines[std::distance(blit_color_keys.begin(), it)];
     }
     blit_color_keys.push_back(key);
+    const VkPipelineRenderingCreateInfo rendering_ci = MakePipelineRenderingCreateInfo(framebuffer);
 
     const std::array stages = MakeStages(*full_screen_vert, *blit_color_to_color_frag);
     const VkPipelineColorBlendAttachmentState blend_attachment{
@@ -1173,7 +1216,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceColorPipeline(const BlitImagePipelineKe
     const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci = GetPipelineInputAssemblyStateCreateInfo(device);
     blit_color_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
-        .pNext = nullptr,
+        .pNext = key.renderpass ? nullptr : &rendering_ci,
         .flags = 0,
         .stageCount = static_cast<u32>(stages.size()),
         .pStages = stages.data(),
@@ -1195,17 +1238,19 @@ VkPipeline BlitImageHelper::FindOrEmplaceColorPipeline(const BlitImagePipelineKe
     return *blit_color_pipelines.back();
 }
 
-VkPipeline BlitImageHelper::FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key) {
+VkPipeline BlitImageHelper::FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key,
+                                                             const Framebuffer* framebuffer) {
     const auto it = std::ranges::find(blit_depth_stencil_keys, key);
     if (it != blit_depth_stencil_keys.end()) {
         return *blit_depth_stencil_pipelines[std::distance(blit_depth_stencil_keys.begin(), it)];
     }
     blit_depth_stencil_keys.push_back(key);
+    const VkPipelineRenderingCreateInfo rendering_ci = MakePipelineRenderingCreateInfo(framebuffer);
     const std::array stages = MakeStages(*full_screen_vert, *blit_depth_stencil_frag);
     const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci = GetPipelineInputAssemblyStateCreateInfo(device);
     blit_depth_stencil_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
-        .pNext = nullptr,
+        .pNext = key.renderpass ? nullptr : &rendering_ci,
         .flags = 0,
         .stageCount = static_cast<u32>(stages.size()),
         .pStages = stages.data(),
@@ -1227,38 +1272,46 @@ VkPipeline BlitImageHelper::FindOrEmplaceDepthStencilPipeline(const BlitImagePip
     return *blit_depth_stencil_pipelines.back();
 }
 
-VkPipeline BlitImageHelper::FindOrEmplaceClearColorPipeline(const BlitImagePipelineKey& key) {
+VkPipeline BlitImageHelper::FindOrEmplaceClearColorPipeline(const BlitImagePipelineKey& key,
+                                                           const Framebuffer* framebuffer) {
     const auto it = std::ranges::find(clear_color_keys, key);
     if (it != clear_color_keys.end()) {
         return *clear_color_pipelines[std::distance(clear_color_keys.begin(), it)];
     }
     clear_color_keys.push_back(key);
+    const VkPipelineRenderingCreateInfo rendering_ci = MakePipelineRenderingCreateInfo(framebuffer);
     const std::array stages = MakeStages(*clear_color_vert, *clear_color_frag);
-    const VkPipelineColorBlendAttachmentState color_blend_attachment_state{
-        .blendEnable = VK_TRUE,
-        .srcColorBlendFactor = VK_BLEND_FACTOR_CONSTANT_COLOR,
-        .dstColorBlendFactor = VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_COLOR,
-        .colorBlendOp = VK_BLEND_OP_ADD,
-        .srcAlphaBlendFactor = VK_BLEND_FACTOR_CONSTANT_ALPHA,
-        .dstAlphaBlendFactor = VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_ALPHA,
-        .alphaBlendOp = VK_BLEND_OP_ADD,
-        .colorWriteMask = VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT |
-                          VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT,
-    };
+    const u32 num_color = framebuffer->NumColorAttachments();
+    constexpr VkColorComponentFlags full_write_mask =
+        VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT | VK_COLOR_COMPONENT_B_BIT |
+        VK_COLOR_COMPONENT_A_BIT;
+    std::array<VkPipelineColorBlendAttachmentState, VideoCommon::NUM_RT> blend_attachments{};
+    for (u32 index = 0; index < num_color; ++index) {
+        blend_attachments[index] = VkPipelineColorBlendAttachmentState{
+            .blendEnable = index == 0 ? VK_TRUE : VK_FALSE,
+            .srcColorBlendFactor = VK_BLEND_FACTOR_CONSTANT_COLOR,
+            .dstColorBlendFactor = VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_COLOR,
+            .colorBlendOp = VK_BLEND_OP_ADD,
+            .srcAlphaBlendFactor = VK_BLEND_FACTOR_CONSTANT_ALPHA,
+            .dstAlphaBlendFactor = VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_ALPHA,
+            .alphaBlendOp = VK_BLEND_OP_ADD,
+            .colorWriteMask = index == 0 ? full_write_mask : VkColorComponentFlags{0},
+        };
+    }
     const VkPipelineColorBlendStateCreateInfo color_blend_state_generic_create_info{
         .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
         .pNext = nullptr,
         .flags = 0,
         .logicOpEnable = VK_FALSE,
         .logicOp = VK_LOGIC_OP_CLEAR,
-        .attachmentCount = 1,
-        .pAttachments = &color_blend_attachment_state,
+        .attachmentCount = num_color,
+        .pAttachments = blend_attachments.data(),
         .blendConstants = {0.0f, 0.0f, 0.0f, 0.0f},
     };
     const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci = GetPipelineInputAssemblyStateCreateInfo(device);
     clear_color_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
-        .pNext = nullptr,
+        .pNext = key.renderpass ? nullptr : &rendering_ci,
         .flags = 0,
         .stageCount = static_cast<u32>(stages.size()),
         .pStages = stages.data(),
@@ -1281,13 +1334,26 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearColorPipeline(const BlitImagePipel
 }
 
 VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline(
-    const BlitDepthStencilPipelineKey& key) {
+    const BlitDepthStencilPipelineKey& key, const Framebuffer* framebuffer) {
     const auto it = std::ranges::find(clear_stencil_keys, key);
     if (it != clear_stencil_keys.end()) {
         return *clear_stencil_pipelines[std::distance(clear_stencil_keys.begin(), it)];
     }
     clear_stencil_keys.push_back(key);
+    const VkPipelineRenderingCreateInfo rendering_ci = MakePipelineRenderingCreateInfo(framebuffer);
     const std::array stages = MakeStages(*clear_color_vert, *clear_stencil_frag);
+    const u32 num_color = framebuffer->NumColorAttachments();
+    std::array<VkPipelineColorBlendAttachmentState, VideoCommon::NUM_RT> blend_attachments{};
+    const VkPipelineColorBlendStateCreateInfo color_blend_ci{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .logicOpEnable = VK_FALSE,
+        .logicOp = VK_LOGIC_OP_CLEAR,
+        .attachmentCount = num_color,
+        .pAttachments = blend_attachments.data(),
+        .blendConstants = {0.0f, 0.0f, 0.0f, 0.0f},
+    };
     const auto stencil = VkStencilOpState{
         .failOp = VK_STENCIL_OP_KEEP,
         .passOp = VK_STENCIL_OP_REPLACE,
@@ -1314,7 +1380,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline(
     const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci = GetPipelineInputAssemblyStateCreateInfo(device);
     clear_stencil_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
-        .pNext = nullptr,
+        .pNext = key.renderpass ? nullptr : &rendering_ci,
         .flags = 0,
         .stageCount = static_cast<u32>(stages.size()),
         .pStages = stages.data(),
@@ -1325,7 +1391,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline(
         .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
         .pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
         .pDepthStencilState = &depth_stencil_ci,
-        .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
+        .pColorBlendState = &color_blend_ci,
         .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
         .layout = *clear_color_pipeline_layout,
         .renderPass = key.renderpass,
@@ -1336,12 +1402,14 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline(
     return *clear_stencil_pipelines.back();
 }
 
-VkPipeline BlitImageHelper::FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPipelineKey& key) {
+VkPipeline BlitImageHelper::FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPipelineKey& key,
+                                                               const Framebuffer* framebuffer) {
     const auto it = std::ranges::find(blit_msaa_color_keys, key);
     if (it != blit_msaa_color_keys.end()) {
         return *blit_msaa_color_pipelines[std::distance(blit_msaa_color_keys.begin(), it)];
     }
     blit_msaa_color_keys.push_back(key);
+    const VkPipelineRenderingCreateInfo rendering_ci = MakePipelineRenderingCreateInfo(framebuffer);
     const std::array stages = MakeStages(*full_screen_vert, *blit_color_msaa_frag);
     const VkPipelineMultisampleStateCreateInfo multisample_ci{
         .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
@@ -1357,7 +1425,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPip
     const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci = GetPipelineInputAssemblyStateCreateInfo(device);
     blit_msaa_color_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
-        .pNext = nullptr,
+        .pNext = key.renderpass ? nullptr : &rendering_ci,
         .flags = 0,
         .stageCount = static_cast<u32>(stages.size()),
         .pStages = stages.data(),
@@ -1379,22 +1447,28 @@ VkPipeline BlitImageHelper::FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPip
     return *blit_msaa_color_pipelines.back();
 }
 
-VkPipeline BlitImageHelper::FindOrEmplaceResolveDepthStencilPipeline(VkRenderPass renderpass,
-                                                                     bool resolve_stencil) {
+VkPipeline BlitImageHelper::FindOrEmplaceResolveDepthStencilPipeline(
+    const Framebuffer* framebuffer, bool resolve_stencil) {
+    const VkRenderPass renderpass = framebuffer->RenderPass();
+    const ResolveDepthStencilPipelineKey key{
+        .renderpass = renderpass,
+        .depth_format = framebuffer->DepthAttachmentFormat(),
+    };
     auto& keys = resolve_stencil ? resolve_depth_stencil_keys : resolve_depth_keys;
     auto& pipelines = resolve_stencil ? resolve_depth_stencil_pipelines : resolve_depth_pipelines;
-    const auto it = std::ranges::find(keys, renderpass);
+    const auto it = std::ranges::find(keys, key);
     if (it != keys.end()) {
         return *pipelines[std::distance(keys.begin(), it)];
     }
-    keys.push_back(renderpass);
+    keys.push_back(key);
+    const VkPipelineRenderingCreateInfo rendering_ci = MakePipelineRenderingCreateInfo(framebuffer);
     const std::array stages =
         MakeStages(*full_screen_vert,
                    resolve_stencil ? *blit_depth_stencil_msaa_frag : *blit_depth_msaa_frag);
     const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci = GetPipelineInputAssemblyStateCreateInfo(device);
     pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
-        .pNext = nullptr,
+        .pNext = renderpass ? nullptr : &rendering_ci,
         .flags = 0,
         .stageCount = static_cast<u32>(stages.size()),
         .pStages = stages.data(),
@@ -1462,25 +1536,29 @@ VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyPipeline(const MSAACopyPipeline
     return *msaa_copy_pipelines.back();
 }
 
-void BlitImageHelper::ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass) {
-    ConvertPipeline(pipeline, renderpass, false);
+void BlitImageHelper::ConvertDepthToColorPipeline(vk::Pipeline& pipeline,
+                                                  const Framebuffer* framebuffer) {
+    ConvertPipeline(pipeline, framebuffer, false);
 }
 
-void BlitImageHelper::ConvertColorToDepthPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass) {
-    ConvertPipeline(pipeline, renderpass, true);
+void BlitImageHelper::ConvertColorToDepthPipeline(vk::Pipeline& pipeline,
+                                                  const Framebuffer* framebuffer) {
+    ConvertPipeline(pipeline, framebuffer, true);
 }
 
-void BlitImageHelper::ConvertPipelineEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
+void BlitImageHelper::ConvertPipelineEx(vk::Pipeline& pipeline, const Framebuffer* framebuffer,
                                         vk::ShaderModule& module, bool single_texture,
                                         bool is_target_depth) {
     if (pipeline) {
         return;
     }
+    const VkRenderPass renderpass = framebuffer->RenderPass();
+    const VkPipelineRenderingCreateInfo rendering_ci = MakePipelineRenderingCreateInfo(framebuffer);
     const std::array stages = MakeStages(*full_screen_vert, *module);
     const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci = GetPipelineInputAssemblyStateCreateInfo(device);
     pipeline = device.GetLogical().CreateGraphicsPipeline(VkGraphicsPipelineCreateInfo{
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
-        .pNext = nullptr,
+        .pNext = renderpass ? nullptr : &rendering_ci,
         .flags = 0,
         .stageCount = static_cast<u32>(stages.size()),
         .pStages = stages.data(),
@@ -1502,28 +1580,32 @@ void BlitImageHelper::ConvertPipelineEx(vk::Pipeline& pipeline, VkRenderPass ren
     });
 }
 
-void BlitImageHelper::ConvertPipelineColorTargetEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
+void BlitImageHelper::ConvertPipelineColorTargetEx(vk::Pipeline& pipeline,
+                                                   const Framebuffer* framebuffer,
                                                    vk::ShaderModule& module) {
-    ConvertPipelineEx(pipeline, renderpass, module, false, false);
+    ConvertPipelineEx(pipeline, framebuffer, module, false, false);
 }
 
-void BlitImageHelper::ConvertPipelineDepthTargetEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
+void BlitImageHelper::ConvertPipelineDepthTargetEx(vk::Pipeline& pipeline,
+                                                   const Framebuffer* framebuffer,
                                                    vk::ShaderModule& module) {
-    ConvertPipelineEx(pipeline, renderpass, module, true, true);
+    ConvertPipelineEx(pipeline, framebuffer, module, true, true);
 }
 
-void BlitImageHelper::ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass,
+void BlitImageHelper::ConvertPipeline(vk::Pipeline& pipeline, const Framebuffer* framebuffer,
                                     bool is_target_depth) {
     if (pipeline) {
         return;
     }
+    const VkRenderPass renderpass = framebuffer->RenderPass();
+    const VkPipelineRenderingCreateInfo rendering_ci = MakePipelineRenderingCreateInfo(framebuffer);
     VkShaderModule frag_shader =
         is_target_depth ? *convert_float_to_depth_frag : *convert_depth_to_float_frag;
     const std::array stages = MakeStages(*full_screen_vert, frag_shader);
     const VkPipelineInputAssemblyStateCreateInfo input_assembly_ci = GetPipelineInputAssemblyStateCreateInfo(device);
     pipeline = device.GetLogical().CreateGraphicsPipeline(VkGraphicsPipelineCreateInfo{
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
-        .pNext = nullptr,
+        .pNext = renderpass ? nullptr : &rendering_ci,
         .flags = 0,
         .stageCount = static_cast<u32>(stages.size()),
         .pStages = stages.data(),

src/video_core/renderer_vulkan/blit_image.h

@@ -33,6 +33,8 @@ struct BlitImagePipelineKey {
 
     VkRenderPass renderpass;
     Tegra::Engines::Fermi2D::Operation operation;
+    std::array<VkFormat, VideoCommon::NUM_RT> color_formats;
+    VkFormat depth_format;
 };
 
 struct BlitDepthStencilPipelineKey {
@@ -43,6 +45,8 @@ struct BlitDepthStencilPipelineKey {
     u8 stencil_mask;
     u32 stencil_compare_mask;
     u32 stencil_ref;
+    std::array<VkFormat, VideoCommon::NUM_RT> color_formats;
+    VkFormat depth_format;
 };
 
 struct MSAACopyPipelineKey {
@@ -58,6 +62,15 @@ struct BlitMSAAPipelineKey {
 
     VkRenderPass renderpass;
     VkSampleCountFlagBits samples;
+    std::array<VkFormat, VideoCommon::NUM_RT> color_formats;
+    VkFormat depth_format;
+};
+
+struct ResolveDepthStencilPipelineKey {
+    constexpr auto operator<=>(const ResolveDepthStencilPipelineKey&) const noexcept = default;
+
+    VkRenderPass renderpass;
+    VkFormat depth_format;
 };
 
 class BlitImageHelper {
@@ -123,31 +136,36 @@ private:
     void ConvertDepthStencil(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
                              ImageView& src_image_view);
 
-    [[nodiscard]] VkPipeline FindOrEmplaceColorPipeline(const BlitImagePipelineKey& key);
+    [[nodiscard]] VkPipeline FindOrEmplaceColorPipeline(const BlitImagePipelineKey& key,
+                                                        const Framebuffer* framebuffer);
 
-    [[nodiscard]] VkPipeline FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key);
+    [[nodiscard]] VkPipeline FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key,
+                                                               const Framebuffer* framebuffer);
 
-    [[nodiscard]] VkPipeline FindOrEmplaceClearColorPipeline(const BlitImagePipelineKey& key);
+    [[nodiscard]] VkPipeline FindOrEmplaceClearColorPipeline(const BlitImagePipelineKey& key,
+                                                             const Framebuffer* framebuffer);
     [[nodiscard]] VkPipeline FindOrEmplaceClearStencilPipeline(
-        const BlitDepthStencilPipelineKey& key);
+        const BlitDepthStencilPipelineKey& key, const Framebuffer* framebuffer);
     [[nodiscard]] VkPipeline FindOrEmplaceMSAACopyPipeline(const MSAACopyPipelineKey& key);
-    [[nodiscard]] VkPipeline FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPipelineKey& key);
-    [[nodiscard]] VkPipeline FindOrEmplaceResolveDepthStencilPipeline(VkRenderPass renderpass,
+    [[nodiscard]] VkPipeline FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPipelineKey& key,
+                                                                const Framebuffer* framebuffer);
+    [[nodiscard]] VkPipeline FindOrEmplaceResolveDepthStencilPipeline(const Framebuffer* framebuffer,
                                                                       bool resolve_stencil);
 
-    void ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass, bool is_target_depth);
+    void ConvertPipeline(vk::Pipeline& pipeline, const Framebuffer* framebuffer,
+                         bool is_target_depth);
 
-    void ConvertDepthToColorPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass);
+    void ConvertDepthToColorPipeline(vk::Pipeline& pipeline, const Framebuffer* framebuffer);
 
-    void ConvertColorToDepthPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass);
+    void ConvertColorToDepthPipeline(vk::Pipeline& pipeline, const Framebuffer* framebuffer);
 
-    void ConvertPipelineEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
+    void ConvertPipelineEx(vk::Pipeline& pipeline, const Framebuffer* framebuffer,
                            vk::ShaderModule& module, bool single_texture, bool is_target_depth);
 
-    void ConvertPipelineColorTargetEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
+    void ConvertPipelineColorTargetEx(vk::Pipeline& pipeline, const Framebuffer* framebuffer,
                                       vk::ShaderModule& module);
 
-    void ConvertPipelineDepthTargetEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
+    void ConvertPipelineDepthTargetEx(vk::Pipeline& pipeline, const Framebuffer* framebuffer,
                                       vk::ShaderModule& module);
 
     const Device& device;
@@ -195,9 +213,9 @@ private:
     std::vector<vk::Pipeline> msaa_copy_pipelines;
     std::vector<BlitMSAAPipelineKey> blit_msaa_color_keys;
     std::vector<vk::Pipeline> blit_msaa_color_pipelines;
-    std::vector<VkRenderPass> resolve_depth_keys;
+    std::vector<ResolveDepthStencilPipelineKey> resolve_depth_keys;
     std::vector<vk::Pipeline> resolve_depth_pipelines;
-    std::vector<VkRenderPass> resolve_depth_stencil_keys;
+    std::vector<ResolveDepthStencilPipelineKey> resolve_depth_stencil_keys;
     std::vector<vk::Pipeline> resolve_depth_stencil_pipelines;
     struct MSAACopyResources {
         u64 tick;

src/video_core/renderer_vulkan/vk_command_pool.cpp

@@ -1,9 +1,13 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <cstddef>
 
 #include "video_core/renderer_vulkan/vk_command_pool.h"
+#include "video_core/renderer_vulkan/vk_master_semaphore.h"
 #include "video_core/vulkan_common/vulkan_device.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
@@ -14,32 +18,52 @@ constexpr size_t COMMAND_BUFFER_POOL_SIZE = 4;
 struct CommandPool::Pool {
     vk::CommandPool handle;
     vk::CommandBuffers cmdbufs;
+    u64 tick;
 };
 
 CommandPool::CommandPool(MasterSemaphore& master_semaphore_, const Device& device_)
-    : ResourcePool(master_semaphore_, COMMAND_BUFFER_POOL_SIZE), device{device_} {}
+    : master_semaphore{master_semaphore_}, device{device_} {}
 
 CommandPool::~CommandPool() = default;
 
-void CommandPool::Allocate(size_t begin, size_t end) {
-    // Command buffers are going to be committed, recorded, executed every single usage cycle.
-    // They are also going to be reset when committed.
+void CommandPool::AllocatePool() {
     Pool& pool = pools.emplace_back();
     pool.handle = device.GetLogical().CreateCommandPool({
         .sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO,
         .pNext = nullptr,
-        .flags =
-            VK_COMMAND_POOL_CREATE_TRANSIENT_BIT | VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT,
+        .flags = VK_COMMAND_POOL_CREATE_TRANSIENT_BIT,
         .queueFamilyIndex = device.GetGraphicsFamily(),
     });
     pool.cmdbufs = pool.handle.Allocate(COMMAND_BUFFER_POOL_SIZE);
+    pool.tick = 0;
+}
+
+void CommandPool::AcquirePool() {
+    if (!pools.empty()) {
+        master_semaphore.Refresh();
+        const u64 gpu_tick = master_semaphore.KnownGpuTick();
+        for (size_t i = 0; i < pools.size(); ++i) {
+            const size_t candidate = (current_pool + 1 + i) % pools.size();
+            if (gpu_tick >= pools[candidate].tick) {
+                current_pool = candidate;
+                current_index = 0;
+                pools[current_pool].handle.Reset();
+                return;
+            }
+        }
+    }
+    AllocatePool();
+    current_pool = pools.size() - 1;
+    current_index = 0;
 }
 
 VkCommandBuffer CommandPool::Commit() {
-    const size_t index = CommitResource();
-    const auto pool_index = index / COMMAND_BUFFER_POOL_SIZE;
-    const auto sub_index = index % COMMAND_BUFFER_POOL_SIZE;
-    return pools[pool_index].cmdbufs[sub_index];
+    if (pools.empty() || current_index >= COMMAND_BUFFER_POOL_SIZE) {
+        AcquirePool();
+    }
+    Pool& pool = pools[current_pool];
+    pool.tick = master_semaphore.CurrentTick();
+    return pool.cmdbufs[current_index++];
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_command_pool.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -6,7 +9,7 @@
 #include <cstddef>
 #include <vector>
 
-#include "video_core/renderer_vulkan/vk_resource_pool.h"
+#include "common/common_types.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
 namespace Vulkan {
@@ -14,20 +17,24 @@ namespace Vulkan {
 class Device;
 class MasterSemaphore;
 
-class CommandPool final : public ResourcePool {
+class CommandPool final {
 public:
     explicit CommandPool(MasterSemaphore& master_semaphore_, const Device& device_);
-    ~CommandPool() override;
-
-    void Allocate(size_t begin, size_t end) override;
+    ~CommandPool();
 
     VkCommandBuffer Commit();
 
 private:
     struct Pool;
 
+    void AllocatePool();
+    void AcquirePool();
+
+    MasterSemaphore& master_semaphore;
     const Device& device;
     std::vector<Pool> pools;
+    size_t current_pool = 0;
+    size_t current_index = 0;
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -290,7 +290,10 @@ GraphicsPipeline::GraphicsPipeline(
         descriptor_update_template =
             builder.CreateTemplate(set_layout, *pipeline_layout, uses_push_descriptor);
 
-        const VkRenderPass render_pass{render_pass_cache.Get(MakeRenderPassKey(key.state, device))};
+        VkRenderPass render_pass{};
+        if (!device.IsKhrDynamicRenderingSupported()) {
+            render_pass = render_pass_cache.Get(MakeRenderPassKey(key.state, device));
+        }
         Validate();
         try {
             MakePipeline(render_pass);
@@ -996,9 +999,47 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
         flags |= VK_PIPELINE_CREATE_CAPTURE_STATISTICS_BIT_KHR;
     }
 
+    const RenderPassKey renderpass_key{MakeRenderPassKey(key.state, device)};
+    std::array<VkFormat, Maxwell::NumRenderTargets> color_attachment_formats{};
+    for (size_t index = 0; index < renderpass_key.color_formats.size(); ++index) {
+        const PixelFormat pixel_format{renderpass_key.color_formats[index]};
+        if (pixel_format == PixelFormat::Invalid) {
+            color_attachment_formats[index] = VK_FORMAT_UNDEFINED;
+            continue;
+        }
+        color_attachment_formats[index] =
+            MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, true, pixel_format).format;
+    }
+    VkFormat depth_attachment_format{VK_FORMAT_UNDEFINED};
+    VkFormat stencil_attachment_format{VK_FORMAT_UNDEFINED};
+    if (renderpass_key.depth_format != PixelFormat::Invalid) {
+        const VkFormat format{
+            MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, true,
+                                       renderpass_key.depth_format)
+                .format};
+        const auto surface_type{VideoCore::Surface::GetFormatType(renderpass_key.depth_format)};
+        if (surface_type == VideoCore::Surface::SurfaceType::Depth ||
+            surface_type == VideoCore::Surface::SurfaceType::DepthStencil) {
+            depth_attachment_format = format;
+        }
+        if (surface_type == VideoCore::Surface::SurfaceType::Stencil ||
+            surface_type == VideoCore::Surface::SurfaceType::DepthStencil) {
+            stencil_attachment_format = format;
+        }
+    }
+    const VkPipelineRenderingCreateInfo rendering_ci{
+        .sType = VK_STRUCTURE_TYPE_PIPELINE_RENDERING_CREATE_INFO,
+        .pNext = nullptr,
+        .viewMask = 0,
+        .colorAttachmentCount = static_cast<u32>(NumAttachments(key.state)),
+        .pColorAttachmentFormats = color_attachment_formats.data(),
+        .depthAttachmentFormat = depth_attachment_format,
+        .stencilAttachmentFormat = stencil_attachment_format,
+    };
+
     pipeline = device.GetLogical().CreateGraphicsPipeline({
         .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
-        .pNext = nullptr,
+        .pNext = device.IsKhrDynamicRenderingSupported() ? &rendering_ci : nullptr,
         .flags = flags,
         .stageCount = static_cast<u32>(shader_stages.size()),
         .pStages = shader_stages.data(),

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -305,12 +305,8 @@ size_t GetTotalPipelineWorkers() {
         std::max<size_t>(static_cast<size_t>(std::thread::hardware_concurrency()), 2ULL) - 1ULL;
 #ifdef __ANDROID__
     const int configured = AndroidSettings::values.pipeline_worker_count.GetValue();
-    const int clamped = std::clamp(configured, 4, 8);
-    const size_t desired = static_cast<size_t>(clamped);
-    if (desired == 0) {
-        return 1ULL;
-    }
-    return std::min(max_core_threads, desired);
+    const size_t desired = static_cast<size_t>(std::max(configured, 1));
+    return std::min<size_t>(max_core_threads, desired);
 #else
     return max_core_threads;
 #endif

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -241,11 +241,13 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) {
     if (!pipeline) {
         return;
     }
-    std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
-    // update engine as channel may be different.
-    pipeline->SetEngine(maxwell3d, gpu_memory);
-    if (!pipeline->Configure(is_indexed))
-        return;
+    {
+        std::scoped_lock lock{buffer_cache.mutex, texture_cache.mutex};
+        pipeline->SetEngine(maxwell3d, gpu_memory);
+        if (!pipeline->Configure(is_indexed)) {
+            return;
+        }
+    }
 
     UpdateDynamicStates();
 
@@ -673,7 +675,6 @@ void RasterizerVulkan::FlushRegion(DAddr addr, u64 size, VideoCommon::CacheType
         texture_cache.DownloadMemory(addr, size);
     }
     if ((True(which & VideoCommon::CacheType::BufferCache))) {
-        std::scoped_lock lock{buffer_cache.mutex};
         buffer_cache.DownloadMemory(addr, size);
     }
     if ((True(which & VideoCommon::CacheType::QueryCache))) {
@@ -771,12 +772,15 @@ bool RasterizerVulkan::OnCPUWrite(DAddr addr, u64 size) {
     return false;
 }
 
+static constexpr bool ENABLE_TEXTURE_CACHE_INVALIDATION_SKIP = true;
+
 void RasterizerVulkan::OnCacheInvalidation(DAddr addr, u64 size) {
     if (addr == 0 || size == 0) {
         return;
     }
 
-    {
+    if (!ENABLE_TEXTURE_CACHE_INVALIDATION_SKIP ||
+        device_memory.IsRegionTextureCached(addr, size)) {
         std::scoped_lock lock{texture_cache.mutex};
         texture_cache.WriteMemory(addr, size);
     }

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -93,13 +93,53 @@ void Scheduler::DispatchWork() {
     }
 }
 
+void Scheduler::BeginDynamicRendering(const Framebuffer* framebuffer, const DeferredClear* clear) {
+    const VkExtent2D render_area = framebuffer->RenderArea();
+    std::array<VkImageView, 9> attachment_views{};
+    const auto& color_views = framebuffer->ColorAttachments();
+    for (size_t index = 0; index < color_views.size(); ++index) {
+        attachment_views[index] = color_views[index];
+    }
+    attachment_views[8] = framebuffer->DepthAttachment();
+    state.renderpass = VkRenderPass{};
+    state.framebuffer = VkFramebuffer{};
+    state.attachment_views = attachment_views;
+    state.color_resolve_views = framebuffer->ColorResolveAttachments();
+    state.color_resolve_modes = framebuffer->ColorResolveModes();
+    state.discards_msaa_color = framebuffer->DiscardsMsaaColor();
+    state.discards_msaa_depth = framebuffer->DiscardsMsaaDepth();
+    state.render_area = render_area;
+    state.num_color = framebuffer->NumColorAttachments();
+    state.has_depth = framebuffer->HasAspectDepthBit();
+    state.has_stencil = framebuffer->HasAspectStencilBit();
+    state.layer_count = framebuffer->NumLayers();
+    state.rendering = true;
+
+    if (GPU::Logging::IsActive() && Settings::values.gpu_log_vulkan_calls.GetValue()) {
+        const std::string render_pass_info =
+            fmt::format("renderArea={}x{}, numImages={}", render_area.width, render_area.height,
+                        framebuffer->NumImages());
+        GPU::Logging::GPULogger::GetInstance().LogRenderPassBegin(render_pass_info);
+    }
+
+    RecordDynamicBegin(clear);
+    num_renderpass_images = framebuffer->NumImages();
+    renderpass_images = framebuffer->Images();
+    renderpass_image_ranges = framebuffer->ImageRanges();
+}
+
 void Scheduler::BeginRenderPassImpl(const Framebuffer* framebuffer, VkRenderPass renderpass,
                                     const VkClearValue* clear_values, u32 clear_value_count) {
-    const VkFramebuffer framebuffer_handle = framebuffer->Handle();
+    if (device.IsKhrDynamicRenderingSupported()) {
+        BeginDynamicRendering(framebuffer, nullptr);
+        return;
+    }
     const VkExtent2D render_area = framebuffer->RenderArea();
+    const VkFramebuffer framebuffer_handle = framebuffer->Handle();
     state.renderpass = renderpass;
     state.framebuffer = framebuffer_handle;
     state.render_area = render_area;
+    state.rendering = true;
 
     if (GPU::Logging::IsActive() && Settings::values.gpu_log_vulkan_calls.GetValue()) {
         const std::string render_pass_info =
@@ -141,6 +181,12 @@ void Scheduler::RealizeDeferredClear() {
     const DeferredClear dc = deferred_clear;
     deferred_clear = {};
 
+    if (device.IsKhrDynamicRenderingSupported()) {
+        EndRenderPass();
+        BeginDynamicRendering(dc.framebuffer, &dc);
+        return;
+    }
+
     std::array<VkClearValue, 9> clear_values{};
     u32 count = 0;
     const RenderPassKey& base = dc.framebuffer->RenderPassKeyBase();
@@ -195,9 +241,25 @@ void Scheduler::RequestRenderpass(const Framebuffer* framebuffer) {
         RealizeDeferredClear();
         return;
     }
+    const VkExtent2D render_area = framebuffer->RenderArea();
+    if (device.IsKhrDynamicRenderingSupported()) {
+        std::array<VkImageView, 9> attachment_views{};
+        const auto& color_views = framebuffer->ColorAttachments();
+        for (size_t index = 0; index < color_views.size(); ++index) {
+            attachment_views[index] = color_views[index];
+        }
+        attachment_views[8] = framebuffer->DepthAttachment();
+        if (state.rendering && attachment_views == state.attachment_views &&
+            render_area.width == state.render_area.width &&
+            render_area.height == state.render_area.height) {
+            return;
+        }
+        EndRenderPass();
+        BeginDynamicRendering(framebuffer, nullptr);
+        return;
+    }
     const VkRenderPass renderpass = framebuffer->RenderPass();
     const VkFramebuffer framebuffer_handle = framebuffer->Handle();
-    const VkExtent2D render_area = framebuffer->RenderArea();
     if (renderpass == state.renderpass && framebuffer_handle == state.framebuffer &&
         render_area.width == state.render_area.width &&
         render_area.height == state.render_area.height) {
@@ -372,6 +434,80 @@ void Scheduler::InvalidateState() {
     state_tracker.InvalidateCommandBufferState();
 }
 
+void Scheduler::RecordDynamicBegin(const DeferredClear* clear) {
+    const std::array<VkImageView, 9> views = state.attachment_views;
+    const std::array<VkImageView, 8> resolve_views = state.color_resolve_views;
+    const std::array<VkResolveModeFlagBits, 8> resolve_modes = state.color_resolve_modes;
+    const u32 num_color = state.num_color;
+    const bool has_depth = state.has_depth;
+    const bool has_stencil = state.has_stencil;
+    const u32 layers = state.layer_count;
+    const VkExtent2D render_area = state.render_area;
+    const u32 color_clear_mask = clear ? clear->color_clear_mask : 0u;
+    const u32 color_discard_mask =
+        clear != nullptr && state.discards_msaa_color ? clear->color_clear_mask : 0u;
+    const std::array<VkClearValue, 8> color_clear_values =
+        clear ? clear->color_values : std::array<VkClearValue, 8>{};
+    const bool ds_clear = clear != nullptr && clear->depth_stencil;
+    const VkClearValue ds_clear_value = clear ? clear->depth_stencil_value : VkClearValue{};
+    const bool ds_discard = state.discards_msaa_depth;
+    Record([views, resolve_views, resolve_modes, num_color, has_depth, has_stencil, layers,
+            render_area, color_clear_mask, color_discard_mask, color_clear_values, ds_clear,
+            ds_clear_value, ds_discard](vk::CommandBuffer cmdbuf) {
+        std::array<VkRenderingAttachmentInfo, VideoCommon::NUM_RT> color_infos{};
+        for (u32 index = 0; index < num_color; ++index) {
+            const bool clear_slot = ((color_clear_mask >> index) & 1u) != 0;
+            const VkImageView resolve_view = resolve_views[index];
+            const bool has_resolve = resolve_view != VK_NULL_HANDLE;
+            const bool discard_slot = has_resolve && ((color_discard_mask >> index) & 1u) != 0;
+            color_infos[index] = VkRenderingAttachmentInfo{
+                .sType = VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO,
+                .pNext = nullptr,
+                .imageView = views[index],
+                .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
+                .resolveMode = has_resolve ? resolve_modes[index] : VK_RESOLVE_MODE_NONE,
+                .resolveImageView = resolve_view,
+                .resolveImageLayout =
+                    has_resolve ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
+                .loadOp = clear_slot ? VK_ATTACHMENT_LOAD_OP_CLEAR : VK_ATTACHMENT_LOAD_OP_LOAD,
+                .storeOp = discard_slot ? VK_ATTACHMENT_STORE_OP_DONT_CARE
+                                        : VK_ATTACHMENT_STORE_OP_STORE,
+                .clearValue = clear_slot ? color_clear_values[index] : VkClearValue{},
+            };
+        }
+        const VkRenderingAttachmentInfo depth_info{
+            .sType = VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO,
+            .pNext = nullptr,
+            .imageView = views[8],
+            .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .resolveMode = VK_RESOLVE_MODE_NONE,
+            .resolveImageView = VK_NULL_HANDLE,
+            .resolveImageLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+            .loadOp = ds_clear ? VK_ATTACHMENT_LOAD_OP_CLEAR : VK_ATTACHMENT_LOAD_OP_LOAD,
+            .storeOp = ds_discard ? VK_ATTACHMENT_STORE_OP_DONT_CARE
+                                  : VK_ATTACHMENT_STORE_OP_STORE,
+            .clearValue = ds_clear ? ds_clear_value : VkClearValue{},
+        };
+        const VkRenderingInfo rendering_info{
+            .sType = VK_STRUCTURE_TYPE_RENDERING_INFO,
+            .pNext = nullptr,
+            .flags = 0,
+            .renderArea =
+                {
+                    .offset = {.x = 0, .y = 0},
+                    .extent = render_area,
+                },
+            .layerCount = layers,
+            .viewMask = 0,
+            .colorAttachmentCount = num_color,
+            .pColorAttachments = color_infos.data(),
+            .pDepthAttachment = has_depth ? &depth_info : nullptr,
+            .pStencilAttachment = has_stencil ? &depth_info : nullptr,
+        };
+        cmdbuf.BeginRendering(rendering_info);
+    });
+}
+
 void Scheduler::EndPendingOperations() {
     query_cache->CounterReset(VideoCommon::QueryType::ZPassPixelCount64);
     EndRenderPass();
@@ -380,7 +516,7 @@ void Scheduler::EndPendingOperations() {
 void Scheduler::EndRenderPass()
     {
         RealizeDeferredClear();
-        if (!state.renderpass) {
+        if (!state.rendering) {
             return;
         }
 
@@ -398,7 +534,8 @@ void Scheduler::EndRenderPass()
         Record([num_images = num_renderpass_images,
                        images = renderpass_images,
                        ranges = renderpass_image_ranges,
-                       has_transform_feedback = device.IsExtTransformFeedbackSupported()](
+                       has_transform_feedback = device.IsExtTransformFeedbackSupported(),
+                       dynamic_rendering = device.IsKhrDynamicRenderingSupported()](
                           vk::CommandBuffer cmdbuf) {
             std::array<VkImageMemoryBarrier, 9> barriers;
             for (size_t i = 0; i < num_images; ++i) {
@@ -435,7 +572,11 @@ void Scheduler::EndRenderPass()
                         .subresourceRange = range,
                 };
             }
-            cmdbuf.EndRenderPass();
+            if (dynamic_rendering) {
+                cmdbuf.EndRendering();
+            } else {
+                cmdbuf.EndRenderPass();
+            }
             cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
                                    VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, vk::PIPELINE_STAGE_GRAPHICS_COMPUTE,
                                    0, nullptr, nullptr, vk::Span(barriers.data(), num_images));
@@ -453,6 +594,9 @@ void Scheduler::EndRenderPass()
         });
 
         state.renderpass = VkRenderPass{};
+        state.framebuffer = VkFramebuffer{};
+        state.attachment_views = {};
+        state.rendering = false;
         num_renderpass_images = 0;
     }
 

src/video_core/renderer_vulkan/vk_scheduler.h

@@ -71,7 +71,7 @@ public:
 
     /// Returns true when a render pass is currently active in the scheduler state.
     bool IsRenderPassActive() const {
-        return state.renderpass != VK_NULL_HANDLE;
+        return state.rendering;
     }
 
     /// Update the pipeline to the current execution context.
@@ -250,8 +250,18 @@ private:
     struct State {
         VkRenderPass renderpass{};
         VkFramebuffer framebuffer{};
+        std::array<VkImageView, 9> attachment_views{};
+        std::array<VkImageView, 8> color_resolve_views{};
+        std::array<VkResolveModeFlagBits, 8> color_resolve_modes{};
         VkExtent2D render_area = {0, 0};
         GraphicsPipeline* graphics_pipeline = nullptr;
+        bool rendering = false;
+        bool discards_msaa_color = false;
+        bool discards_msaa_depth = false;
+        u32 num_color = 0;
+        bool has_depth = false;
+        bool has_stencil = false;
+        u32 layer_count = 1;
         bool is_rescaling = false;
         bool rescaling_defined = false;
         bool needs_state_enable_refresh = false;
@@ -269,6 +279,9 @@ private:
     void BeginRenderPassImpl(const Framebuffer* framebuffer, VkRenderPass renderpass,
                              const VkClearValue* clear_values, u32 clear_value_count);
 
+    /// Begins a dynamic rendering pass, optionally realizing a deferred clear via load ops.
+    void BeginDynamicRendering(const Framebuffer* framebuffer, const DeferredClear* clear);
+
     /// If a deferred clear is pending.
     void RealizeDeferredClear();
 
@@ -282,6 +295,8 @@ private:
 
     void EndPendingOperations();
 
+    void RecordDynamicBegin(const DeferredClear* clear);
+
     void EndRenderPass();
 
     void AcquireNewChunk();

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -54,6 +54,7 @@ using VideoCore::Surface::SurfaceType;
 namespace {
 constexpr bool ENABLE_MSAA_RESOLVE_CONSUME = true;
 constexpr bool ENABLE_MSAA_COLOR_DISCARD = true;
+constexpr bool ENABLE_MSAA_DEPTH_DISCARD = true;
 
 constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
     if (color == std::array<float, 4>{0, 0, 0, 0}) {
@@ -1041,6 +1042,32 @@ VkImageView TextureCacheRuntime::GetOrCreateResolveShadow(VkImage msaa_image, Vk
     shadow.extent = extent;
     shadow.layers = layers;
     shadow.up_to_date = true;
+    if (device.IsKhrDynamicRenderingSupported()) {
+        scheduler.RecordWithUploadBuffer(
+            [image = *shadow.image, layers](vk::CommandBuffer, vk::CommandBuffer upload_cmdbuf) {
+                const VkImageMemoryBarrier barrier{
+                    .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+                    .pNext = nullptr,
+                    .srcAccessMask = 0,
+                    .dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
+                    .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+                    .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+                    .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                    .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                    .image = image,
+                    .subresourceRange{
+                        .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                        .baseMipLevel = 0,
+                        .levelCount = 1,
+                        .baseArrayLayer = 0,
+                        .layerCount = layers,
+                    },
+                };
+                upload_cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
+                                              VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, 0,
+                                              barrier);
+            });
+    }
     return *shadow.view;
 }
 
@@ -1353,7 +1380,7 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
 }
 
 void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view) {
-    if (!dst->RenderPass()) {
+    if (!dst->RenderPass() && !device.IsKhrDynamicRenderingSupported()) {
         return;
     }
 
@@ -2683,7 +2710,7 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM
           .height = key.size.height,
       }} {
     CreateFramebuffer(runtime, color_buffers, depth_buffer, key.is_rescaled);
-    if (runtime.device.HasDebuggingToolAttached()) {
+    if (runtime.device.HasDebuggingToolAttached() && framebuffer) {
         framebuffer.SetObjectNameEXT(VideoCommon::Name(key).c_str());
     }
 }
@@ -2726,6 +2753,12 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
         image_ranges[num_images] = MakeSubresourceRange(color_buffer);
         rt_map[index] = num_images;
         samples = color_buffer->Samples();
+        color_attachments[index] = color_buffer->RenderTarget();
+        color_attachment_formats[index] =
+            MaxwellToVK::SurfaceFormat(runtime.device, FormatType::Optimal, true,
+                                       color_buffer->format)
+                .format;
+        num_color_attachments = static_cast<u32>(index + 1);
         ++num_images;
     }
     const size_t num_colors = attachments.size();
@@ -2741,6 +2774,11 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
         const VkImageSubresourceRange subresource_range = MakeSubresourceRange(depth_buffer);
         image_ranges[num_images] = subresource_range;
         samples = depth_buffer->Samples();
+        depth_attachment = depth_buffer->RenderTarget();
+        depth_attachment_format =
+            MaxwellToVK::SurfaceFormat(runtime.device, FormatType::Optimal, true,
+                                       depth_buffer->format)
+                .format;
         ++num_images;
         has_depth = (subresource_range.aspectMask & VK_IMAGE_ASPECT_DEPTH_BIT) != 0;
         has_stencil = (subresource_range.aspectMask & VK_IMAGE_ASPECT_STENCIL_BIT) != 0;
@@ -2754,8 +2792,9 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
 
     discard_msaa_color =
         ENABLE_MSAA_RESOLVE_CONSUME && ENABLE_MSAA_COLOR_DISCARD && do_resolve_color;
+    discard_msaa_depth = ENABLE_MSAA_RESOLVE_CONSUME && ENABLE_MSAA_DEPTH_DISCARD &&
+                         samples != VK_SAMPLE_COUNT_1_BIT && has_depth && runtime.device.IsTiler();
 
-    renderpass = runtime.render_pass_cache.Get(renderpass_key);
     render_pass_key = renderpass_key;
     render_pass_cache = &runtime.render_pass_cache;
     render_area.width = (std::min)(render_area.width, width);
@@ -2770,10 +2809,15 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
             }
             const VkFormat vk_format =
                 MaxwellToVK::SurfaceFormat(runtime.device, FormatType::Optimal, true, format).format;
+            color_resolve_modes[index] = VideoCore::Surface::IsPixelFormatInteger(format)
+                                             ? VK_RESOLVE_MODE_SAMPLE_ZERO_BIT
+                                             : VK_RESOLVE_MODE_AVERAGE_BIT;
             if (ENABLE_MSAA_RESOLVE_CONSUME) {
                 const VkImage msaa_image = images[rt_map[index]];
-                attachments.push_back(runtime.GetOrCreateResolveShadow(msaa_image, vk_format,
-                                                                       render_area, layers));
+                const VkImageView shadow_view = runtime.GetOrCreateResolveShadow(
+                    msaa_image, vk_format, render_area, layers);
+                color_resolve_attachments[index] = shadow_view;
+                attachments.push_back(shadow_view);
                 continue;
             }
             VkImageCreateInfo resolve_ci{
@@ -2812,6 +2856,33 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
                         .layerCount = layers,
                     },
                 });
+            if (runtime.device.IsKhrDynamicRenderingSupported()) {
+                runtime.scheduler.RecordWithUploadBuffer(
+                    [image = *resolve_image, layers](vk::CommandBuffer, vk::CommandBuffer upload_cmdbuf) {
+                        const VkImageMemoryBarrier barrier{
+                            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+                            .pNext = nullptr,
+                            .srcAccessMask = 0,
+                            .dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
+                            .oldLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+                            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+                            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+                            .image = image,
+                            .subresourceRange{
+                                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                                .baseMipLevel = 0,
+                                .levelCount = 1,
+                                .baseArrayLayer = 0,
+                                .layerCount = layers,
+                            },
+                        };
+                        upload_cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
+                                                      VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
+                                                      0, barrier);
+                    });
+            }
+            color_resolve_attachments[index] = *resolve_view;
             attachments.push_back(*resolve_view);
             resolve_images.push_back(std::move(resolve_image));
             resolve_image_views.push_back(std::move(resolve_view));
@@ -2819,6 +2890,11 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
     }
 
     num_color_buffers = static_cast<u32>(num_colors);
+    layer_count = static_cast<u32>((std::max)(num_layers, 1));
+    if (runtime.device.IsKhrDynamicRenderingSupported()) {
+        return;
+    }
+    renderpass = runtime.render_pass_cache.Get(renderpass_key);
     framebuffer = runtime.device.GetLogical().CreateFramebuffer({
         .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
         .pNext = nullptr,
@@ -2828,7 +2904,7 @@ void Framebuffer::CreateFramebuffer(TextureCacheRuntime& runtime,
         .pAttachments = attachments.data(),
         .width = render_area.width,
         .height = render_area.height,
-        .layers = static_cast<u32>((std::max)(num_layers, 1)),
+        .layers = layer_count,
     });
 }
 
@@ -2844,6 +2920,53 @@ VkRenderPass Framebuffer::RenderPassVariant(u32 color_clear_mask, bool depth_ste
     return render_pass_cache->Get(key);
 }
 
+void Framebuffer::BeginRendering(vk::CommandBuffer cmdbuf) const {
+    std::array<VkRenderingAttachmentInfo, NUM_RT> color_attachment_infos{};
+    for (size_t index = 0; index < num_color_attachments; ++index) {
+        color_attachment_infos[index] = VkRenderingAttachmentInfo{
+            .sType = VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO,
+            .pNext = nullptr,
+            .imageView = color_attachments[index],
+            .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .resolveMode = VK_RESOLVE_MODE_NONE,
+            .resolveImageView = VK_NULL_HANDLE,
+            .resolveImageLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+            .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
+            .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
+            .clearValue = {},
+        };
+    }
+    const VkRenderingAttachmentInfo depth_attachment_info{
+        .sType = VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO,
+        .pNext = nullptr,
+        .imageView = depth_attachment,
+        .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
+        .resolveMode = VK_RESOLVE_MODE_NONE,
+        .resolveImageView = VK_NULL_HANDLE,
+        .resolveImageLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+        .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
+        .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
+        .clearValue = {},
+    };
+    const VkRenderingInfo rendering_info{
+        .sType = VK_STRUCTURE_TYPE_RENDERING_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .renderArea =
+            {
+                .offset = {.x = 0, .y = 0},
+                .extent = render_area,
+            },
+        .layerCount = layer_count,
+        .viewMask = 0,
+        .colorAttachmentCount = num_color_attachments,
+        .pColorAttachments = color_attachment_infos.data(),
+        .pDepthAttachment = has_depth ? &depth_attachment_info : nullptr,
+        .pStencilAttachment = has_stencil ? &depth_attachment_info : nullptr,
+    };
+    cmdbuf.BeginRendering(rendering_info);
+}
+
 void TextureCacheRuntime::AccelerateImageUpload(
     Image& image, const StagingBufferRef& map,
     std::span<const VideoCommon::SwizzleParameters> swizzles,

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -178,6 +178,8 @@ public:
                            std::span<ImageView*, NUM_RT> color_buffers, ImageView* depth_buffer,
                            bool is_rescaled = false);
 
+    void BeginRendering(vk::CommandBuffer cmdbuf) const;
+
     [[nodiscard]] VkFramebuffer Handle() const noexcept {
         return *framebuffer;
     }
@@ -193,6 +195,39 @@ public:
     [[nodiscard]] VkRenderPass RenderPassVariant(u32 color_clear_mask, bool depth_stencil_clear,
                                                  u32 color_discard_mask) const;
 
+    [[nodiscard]] u32 NumColorAttachments() const noexcept {
+        return num_color_attachments;
+    }
+
+    [[nodiscard]] const std::array<VkImageView, NUM_RT>& ColorAttachments() const noexcept {
+        return color_attachments;
+    }
+
+    [[nodiscard]] const std::array<VkFormat, NUM_RT>& ColorAttachmentFormats() const noexcept {
+        return color_attachment_formats;
+    }
+
+    [[nodiscard]] const std::array<VkImageView, NUM_RT>& ColorResolveAttachments() const noexcept {
+        return color_resolve_attachments;
+    }
+
+    [[nodiscard]] const std::array<VkResolveModeFlagBits, NUM_RT>& ColorResolveModes()
+        const noexcept {
+        return color_resolve_modes;
+    }
+
+    [[nodiscard]] VkImageView DepthAttachment() const noexcept {
+        return depth_attachment;
+    }
+
+    [[nodiscard]] VkFormat DepthAttachmentFormat() const noexcept {
+        return depth_attachment_format;
+    }
+
+    [[nodiscard]] u32 NumLayers() const noexcept {
+        return layer_count;
+    }
+
     [[nodiscard]] VkExtent2D RenderArea() const noexcept {
         return render_area;
     }
@@ -245,6 +280,10 @@ public:
         return discard_msaa_color;
     }
 
+    [[nodiscard]] bool DiscardsMsaaDepth() const noexcept {
+        return discard_msaa_depth;
+    }
+
 private:
     vk::Framebuffer framebuffer;
     VkRenderPass renderpass{};
@@ -255,6 +294,14 @@ private:
     std::array<VkImage, 9> images{};
     std::array<VkImageSubresourceRange, 9> image_ranges{};
     std::array<size_t, NUM_RT> rt_map{};
+    std::array<VkImageView, NUM_RT> color_attachments{};
+    std::array<VkFormat, NUM_RT> color_attachment_formats{};
+    std::array<VkImageView, NUM_RT> color_resolve_attachments{};
+    std::array<VkResolveModeFlagBits, NUM_RT> color_resolve_modes{};
+    VkImageView depth_attachment{};
+    VkFormat depth_attachment_format = VK_FORMAT_UNDEFINED;
+    u32 num_color_attachments = 0;
+    u32 layer_count = 1;
     bool has_depth{};
     bool has_stencil{};
     bool is_rescaled{};
@@ -263,6 +310,7 @@ private:
     RenderPassKey render_pass_key{};
     RenderPassCache* render_pass_cache{nullptr};
     bool discard_msaa_color{};
+    bool discard_msaa_depth{};
 };
 
 class Image : public VideoCommon::ImageBase {

src/video_core/texture_cache/texture_cache.h

@@ -2308,6 +2308,7 @@ void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) {
     if (False(image.flags & ImageFlagBits::Sparse)) {
         if (image.cpu_addr < ~(1ULL << 40)) {
             device_memory.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, 1);
+            device_memory.UpdateTexturePagesCount(image.cpu_addr, image.guest_size_bytes, 1);
         }
         return;
     }
@@ -2320,12 +2321,14 @@ void TextureCache<P>::TrackImage(ImageBase& image, ImageId image_id) {
             const DAddr cpu_addr = map.cpu_addr;
             const std::size_t size = map.size;
             device_memory.UpdatePagesCachedCount(cpu_addr, size, 1);
+            device_memory.UpdateTexturePagesCount(cpu_addr, size, 1);
         }
         return;
     }
     ForEachSparseSegment(image,
                          [this]([[maybe_unused]] GPUVAddr gpu_addr, DAddr cpu_addr, size_t size) {
                              device_memory.UpdatePagesCachedCount(cpu_addr, size, 1);
+                             device_memory.UpdateTexturePagesCount(cpu_addr, size, 1);
                          });
 }
 
@@ -2336,6 +2339,7 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) {
     if (False(image.flags & ImageFlagBits::Sparse)) {
         if (image.cpu_addr < ~(1ULL << 40)) {
             device_memory.UpdatePagesCachedCount(image.cpu_addr, image.guest_size_bytes, -1);
+            device_memory.UpdateTexturePagesCount(image.cpu_addr, image.guest_size_bytes, -1);
         }
         return;
     }
@@ -2348,6 +2352,7 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) {
         const DAddr cpu_addr = map.cpu_addr;
         const std::size_t size = map.size;
         device_memory.UpdatePagesCachedCount(cpu_addr, size, -1);
+        device_memory.UpdateTexturePagesCount(cpu_addr, size, -1);
     }
 }
 

src/video_core/texture_cache/texture_cache_base.h

@@ -257,7 +257,7 @@ public:
     /// Prepare an image to be used
     void PrepareImage(ImageId image_id, bool is_modification, bool invalidate);
 
-    std::recursive_mutex mutex;
+    std::mutex mutex;
 
 private:
     /// Iterate over all page indices in a range

src/video_core/vulkan_common/vulkan_device.cpp

@@ -1373,6 +1373,11 @@ void Device::RemoveUnsuitableExtensions() {
     extensions.maintenance3 = loaded_extensions.contains(VK_KHR_MAINTENANCE_3_EXTENSION_NAME);
     RemoveExtensionIfUnsuitable(extensions.maintenance3, VK_KHR_MAINTENANCE_3_EXTENSION_NAME);
 
+    // VK_KHR_dynamic_rendering
+    extensions.dynamic_rendering = features.dynamic_rendering.dynamicRendering;
+    RemoveExtensionFeatureIfUnsuitable(extensions.dynamic_rendering, features.dynamic_rendering,
+                                       VK_KHR_DYNAMIC_RENDERING_EXTENSION_NAME);
+
     // VK_KHR_maintenance4
     extensions.maintenance4 = features.maintenance4.maintenance4;
     RemoveExtensionFeatureIfUnsuitable(extensions.maintenance4, features.maintenance4,

src/video_core/vulkan_common/vulkan_device.h

@@ -43,6 +43,7 @@ VK_DEFINE_HANDLE(VmaAllocator)
     FEATURE(EXT, ShaderDemoteToHelperInvocation, SHADER_DEMOTE_TO_HELPER_INVOCATION,               \
             shader_demote_to_helper_invocation)                                                    \
     FEATURE(EXT, SubgroupSizeControl, SUBGROUP_SIZE_CONTROL, subgroup_size_control)                \
+    FEATURE(KHR, DynamicRendering, DYNAMIC_RENDERING, dynamic_rendering)                           \
     FEATURE(KHR, Maintenance4, MAINTENANCE_4, maintenance4)                                        \
     FEATURE(KHR, Synchronization2, SYNCHRONIZATION_2, synchronization2)
 
@@ -941,6 +942,13 @@ FN_MAX_LIMIT_LIST
         return extensions.maintenance3;
     }
 
+    static constexpr bool ENABLE_DYNAMIC_RENDERING = false;
+
+    /// Returns true if the device supports VK_KHR_dynamic_rendering.
+    bool IsKhrDynamicRenderingSupported() const {
+        return ENABLE_DYNAMIC_RENDERING && extensions.dynamic_rendering;
+    }
+
     /// Returns true if the device supports VK_KHR_maintenance4.
     bool IsKhrMaintenance4Supported() const {
         return extensions.maintenance4;

src/video_core/vulkan_common/vulkan_wrapper.cpp

@@ -92,6 +92,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
     X(vkCmdBeginConditionalRenderingEXT);
     X(vkCmdBeginQuery);
     X(vkCmdBeginRenderPass);
+    X(vkCmdBeginRendering);
     X(vkCmdBeginTransformFeedbackEXT);
     X(vkCmdBeginDebugUtilsLabelEXT);
     X(vkCmdBindDescriptorSets);
@@ -119,6 +120,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
     X(vkCmdEndConditionalRenderingEXT);
     X(vkCmdEndQuery);
     X(vkCmdEndRenderPass);
+    X(vkCmdEndRendering);
     X(vkCmdEndTransformFeedbackEXT);
     X(vkCmdEndDebugUtilsLabelEXT);
     X(vkCmdFillBuffer);
@@ -230,6 +232,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
     X(vkMapMemory);
     X(vkQueueSubmit);
     X(vkQueueSubmit2);
+    X(vkResetCommandPool);
     X(vkResetFences);
     X(vkResetQueryPool);
     X(vkSetDebugUtilsObjectNameEXT);
@@ -257,6 +260,12 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
         Proc(dld.vkCmdDrawIndexedIndirectCount, dld, "vkCmdDrawIndexedIndirectCountKHR", device);
     }
 
+    // Support for dynamic rendering is optional until Vulkan 1.3
+    if (!dld.vkCmdBeginRendering) {
+        Proc(dld.vkCmdBeginRendering, dld, "vkCmdBeginRenderingKHR", device);
+        Proc(dld.vkCmdEndRendering, dld, "vkCmdEndRenderingKHR", device);
+    }
+
     // Synchronization2 is core in Vulkan 1.3, otherwise requires VK_KHR_synchronization2
     if (!dld.vkCmdPipelineBarrier2) {
         Proc(dld.vkCmdPipelineBarrier2, dld, "vkCmdPipelineBarrier2KHR", device);

src/video_core/vulkan_common/vulkan_wrapper.h

@@ -208,6 +208,7 @@ struct DeviceDispatch : InstanceDispatch {
     PFN_vkCmdBeginDebugUtilsLabelEXT vkCmdBeginDebugUtilsLabelEXT{};
     PFN_vkCmdBeginQuery vkCmdBeginQuery{};
     PFN_vkCmdBeginRenderPass vkCmdBeginRenderPass{};
+    PFN_vkCmdBeginRendering vkCmdBeginRendering{};
     PFN_vkCmdBeginTransformFeedbackEXT vkCmdBeginTransformFeedbackEXT{};
     PFN_vkCmdBindDescriptorSets vkCmdBindDescriptorSets{};
     PFN_vkCmdBindIndexBuffer vkCmdBindIndexBuffer{};
@@ -236,6 +237,7 @@ struct DeviceDispatch : InstanceDispatch {
     PFN_vkCmdEndDebugUtilsLabelEXT vkCmdEndDebugUtilsLabelEXT{};
     PFN_vkCmdEndQuery vkCmdEndQuery{};
     PFN_vkCmdEndRenderPass vkCmdEndRenderPass{};
+    PFN_vkCmdEndRendering vkCmdEndRendering{};
     PFN_vkCmdEndTransformFeedbackEXT vkCmdEndTransformFeedbackEXT{};
     PFN_vkCmdFillBuffer vkCmdFillBuffer{};
     PFN_vkCmdPipelineBarrier vkCmdPipelineBarrier{};
@@ -346,6 +348,7 @@ struct DeviceDispatch : InstanceDispatch {
     PFN_vkMapMemory vkMapMemory{};
     PFN_vkQueueSubmit vkQueueSubmit{};
     PFN_vkQueueSubmit2 vkQueueSubmit2{};
+    PFN_vkResetCommandPool vkResetCommandPool{};
     PFN_vkResetFences vkResetFences{};
     PFN_vkResetQueryPool vkResetQueryPool{};
     PFN_vkSetDebugUtilsObjectNameEXT vkSetDebugUtilsObjectNameEXT{};
@@ -925,6 +928,10 @@ public:
     CommandBuffers Allocate(std::size_t num_buffers,
                             VkCommandBufferLevel level = VK_COMMAND_BUFFER_LEVEL_PRIMARY) const;
 
+    void Reset(VkCommandPoolResetFlags flags = 0) const {
+        Check(dld->vkResetCommandPool(owner, handle, flags));
+    }
+
     /// Set object name.
     void SetObjectNameEXT(const char* name) const;
 };
@@ -1186,6 +1193,14 @@ public:
         dld->vkCmdEndRenderPass(handle);
     }
 
+    void BeginRendering(const VkRenderingInfo& rendering_info) const noexcept {
+        dld->vkCmdBeginRendering(handle, &rendering_info);
+    }
+
+    void EndRendering() const noexcept {
+        dld->vkCmdEndRendering(handle);
+    }
+
     void BeginQuery(VkQueryPool query_pool, u32 query, VkQueryControlFlags flags) const noexcept {
         dld->vkCmdBeginQuery(handle, query_pool, query, flags);
     }