early-access version 3280

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 3279.
+This is the source code for early-access 3280.
 
 ## Legal Notice
 

src/common/range_map.h

@@ -60,7 +60,7 @@ private:
     using ConstIteratorType = typename MapType::const_iterator;
 
     size_t ContinousSizeInternal(KeyT address) const {
-        const auto it = GetFirstElemnentBeforeOrOn(address);
+        const auto it = GetFirstElementBeforeOrOn(address);
         if (it == container.end() || it->second == null_value) {
             return 0;
         }
@@ -72,14 +72,14 @@ private:
     }
 
     ValueT GetValueInternal(KeyT address) const {
-        const auto it = GetFirstElemnentBeforeOrOn(address);
+        const auto it = GetFirstElementBeforeOrOn(address);
         if (it == container.end()) {
             return null_value;
         }
         return it->second;
     }
 
-    ConstIteratorType GetFirstElemnentBeforeOrOn(KeyT address) const {
+    ConstIteratorType GetFirstElementBeforeOrOn(KeyT address) const {
         auto it = container.lower_bound(address);
         if (it == container.begin()) {
             return it;

src/common/settings.h

@@ -532,6 +532,7 @@ struct Values {
     Setting<bool> reporting_services{false, "reporting_services"};
     Setting<bool> quest_flag{false, "quest_flag"};
     Setting<bool> disable_macro_jit{false, "disable_macro_jit"};
+    Setting<bool> disable_macro_hle{false, "disable_macro_hle"};
     Setting<bool> extended_logging{false, "extended_logging"};
     Setting<bool> use_debug_asserts{false, "use_debug_asserts"};
     Setting<bool> use_auto_stub{false, "use_auto_stub"};

src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp

@@ -143,6 +143,9 @@ void EmitGetAttribute(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr, Scal
     case IR::Attribute::BaseVertex:
         ctx.Add("MOV.F {}.x,{}.baseVertex;", inst, ctx.attrib_name);
         break;
+    case IR::Attribute::DrawID:
+        ctx.Add("MOV.F {}.x,{}.draw.id;", inst, ctx.attrib_name);
+        break;
     case IR::Attribute::FrontFace:
         ctx.Add("CMP.F {}.x,{}.facing.x,0,-1;", inst, ctx.attrib_name);
         break;
@@ -168,6 +171,9 @@ void EmitGetAttributeU32(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr, S
     case IR::Attribute::BaseVertex:
         ctx.Add("MOV.S {}.x,{}.baseVertex;", inst, ctx.attrib_name);
         break;
+    case IR::Attribute::DrawID:
+        ctx.Add("MOV.S {}.x,{}.draw.id;", inst, ctx.attrib_name);
+        break;
     default:
         throw NotImplementedException("Get U32 attribute {}", attr);
     }

src/shader_recompiler/backend/glsl/emit_glsl_context_get_set.cpp

@@ -240,6 +240,9 @@ void EmitGetAttribute(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr,
     case IR::Attribute::BaseVertex:
         ctx.AddF32("{}=itof(gl_BaseVertex);", inst);
         break;
+    case IR::Attribute::DrawID:
+        ctx.AddF32("{}=itof(gl_DrawID);", inst);
+        break;
     default:
         throw NotImplementedException("Get attribute {}", attr);
     }
@@ -262,6 +265,9 @@ void EmitGetAttributeU32(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr, s
     case IR::Attribute::BaseVertex:
         ctx.AddU32("{}=uint(gl_BaseVertex);", inst);
         break;
+    case IR::Attribute::DrawID:
+        ctx.AddU32("{}=uint(gl_DrawID);", inst);
+        break;
     default:
         throw NotImplementedException("Get U32 attribute {}", attr);
     }

src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp

@@ -343,6 +343,8 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, Id vertex) {
         return ctx.OpBitcast(ctx.F32[1], ctx.OpLoad(ctx.U32[1], ctx.base_instance));
     case IR::Attribute::BaseVertex:
         return ctx.OpBitcast(ctx.F32[1], ctx.OpLoad(ctx.U32[1], ctx.base_vertex));
+    case IR::Attribute::DrawID:
+        return ctx.OpBitcast(ctx.F32[1], ctx.OpLoad(ctx.U32[1], ctx.draw_index));
     case IR::Attribute::FrontFace:
         return ctx.OpSelect(ctx.F32[1], ctx.OpLoad(ctx.U1, ctx.front_face),
                             ctx.OpBitcast(ctx.F32[1], ctx.Const(std::numeric_limits<u32>::max())),
@@ -388,6 +390,8 @@ Id EmitGetAttributeU32(EmitContext& ctx, IR::Attribute attr, Id) {
         return ctx.OpLoad(ctx.U32[1], ctx.base_instance);
     case IR::Attribute::BaseVertex:
         return ctx.OpLoad(ctx.U32[1], ctx.base_vertex);
+    case IR::Attribute::DrawID:
+        return ctx.OpLoad(ctx.U32[1], ctx.draw_index);
     default:
         throw NotImplementedException("Read U32 attribute {}", attr);
     }

src/shader_recompiler/backend/spirv/spirv_emit_context.cpp

@@ -1402,6 +1402,9 @@ void EmitContext::DefineInputs(const IR::Program& program) {
     } else if (loads[IR::Attribute::BaseVertex]) {
         base_vertex = DefineInput(*this, U32[1], true, spv::BuiltIn::BaseVertex);
     }
+    if (loads[IR::Attribute::DrawID]) {
+        draw_index = DefineInput(*this, U32[1], true, spv::BuiltIn::DrawIndex);
+    }
     if (loads[IR::Attribute::FrontFace]) {
         front_face = DefineInput(*this, U1, true, spv::BuiltIn::FrontFacing);
     }

src/shader_recompiler/backend/spirv/spirv_emit_context.h

@@ -218,6 +218,7 @@ public:
     Id base_instance{};
     Id vertex_id{};
     Id vertex_index{};
+    Id draw_index{};
     Id base_vertex{};
     Id front_face{};
     Id point_coord{};

src/shader_recompiler/frontend/ir/attribute.cpp

@@ -450,6 +450,8 @@ std::string NameOf(Attribute attribute) {
         return "BaseInstance";
     case Attribute::BaseVertex:
         return "BaseVertex";
+    case Attribute::DrawID:
+        return "DrawID";
     }
     return fmt::format("<reserved attribute {}>", static_cast<int>(attribute));
 }

src/shader_recompiler/frontend/ir/attribute.h

@@ -223,6 +223,7 @@ enum class Attribute : u64 {
     // Implementation attributes
     BaseInstance = 256,
     BaseVertex = 257,
+    DrawID = 258,
 };
 
 constexpr size_t NUM_GENERICS = 32;

src/shader_recompiler/ir_opt/constant_propagation_pass.cpp

@@ -518,6 +518,7 @@ void FoldBitCast(IR::Inst& inst, IR::Opcode reverse) {
             case IR::Attribute::VertexId:
             case IR::Attribute::BaseVertex:
             case IR::Attribute::BaseInstance:
+            case IR::Attribute::DrawID:
                 break;
             default:
                 return;
@@ -665,6 +666,8 @@ void FoldConstBuffer(Environment& env, IR::Block& block, IR::Inst& inst) {
             return IR::Attribute::BaseInstance;
         case ReplaceConstant::BaseVertex:
             return IR::Attribute::BaseVertex;
+        case ReplaceConstant::DrawID:
+            return IR::Attribute::DrawID;
         default:
             throw NotImplementedException("Not implemented replacement variable {}", *replacement);
         }

src/shader_recompiler/shader_info.h

@@ -19,6 +19,7 @@ namespace Shader {
 enum class ReplaceConstant : u32 {
     BaseInstance,
     BaseVertex,
+    DrawID,
 };
 
 enum class TextureType : u32 {

src/video_core/engines/maxwell_3d.cpp

@@ -220,9 +220,6 @@ void Maxwell3D::ProcessMacro(u32 method, const u32* base_start, u32 amount, bool
 }
 
 void Maxwell3D::RefreshParametersImpl() {
-    if (!Settings::IsGPULevelHigh()) {
-        return;
-    }
     size_t current_index = 0;
     for (auto& segment : macro_segments) {
         if (segment.first == 0) {
@@ -448,9 +445,11 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
     case MAXWELL3D_REG_INDEX(const_buffer.buffer) + 15:
         ProcessCBMultiData(base_start, amount);
         break;
-    case MAXWELL3D_REG_INDEX(inline_data):
+    case MAXWELL3D_REG_INDEX(inline_data): {
+        ASSERT(methods_pending == amount);
         upload_state.ProcessData(base_start, amount);
         return;
+    }
     default:
         for (u32 i = 0; i < amount; i++) {
             CallMethod(method, base_start[i], methods_pending - i <= 1);
@@ -537,7 +536,7 @@ void Maxwell3D::ProcessQueryGet() {
 void Maxwell3D::ProcessQueryCondition() {
     const GPUVAddr condition_address{regs.render_enable.Address()};
     switch (regs.render_enable_override) {
-    case Regs::RenderEnable::Override::AlwaysRender: {
+    case Regs::RenderEnable::Override::AlwaysRender:
         execute_on = true;
         break;
     case Regs::RenderEnable::Override::NeverRender:
@@ -586,7 +585,6 @@ void Maxwell3D::ProcessQueryCondition() {
         break;
     }
     }
-    }
 }
 
 void Maxwell3D::ProcessCounterReset() {

src/video_core/engines/maxwell_3d.h

@@ -1218,12 +1218,12 @@ public:
 
         struct Window {
             union {
-                u32 raw_1;
+                u32 raw_x;
                 BitField<0, 16, u32> x_min;
                 BitField<16, 16, u32> x_max;
             };
             union {
-                u32 raw_2;
+                u32 raw_y;
                 BitField<0, 16, u32> y_min;
                 BitField<16, 16, u32> y_max;
             };
@@ -3034,6 +3034,7 @@ public:
     enum class HLEReplacementAttributeType : u32 {
         BaseVertex = 0x0,
         BaseInstance = 0x1,
+        DrawID = 0x2,
     };
 
     void SetHLEReplacementAttributeType(u32 bank, u32 offset, HLEReplacementAttributeType name);
@@ -3112,6 +3113,10 @@ public:
     /// Handles a write to the CB_BIND register.
     void ProcessCBBind(size_t stage_index);
 
+    /// Handles a write to the CB_DATA[i] register.
+    void ProcessCBData(u32 value);
+    void ProcessCBMultiData(const u32* start_base, u32 amount);
+
 private:
     void InitializeRegisterDefaults();
 
@@ -3163,10 +3168,6 @@ private:
     /// Handles writes to syncing register.
     void ProcessSyncPoint();
 
-    /// Handles a write to the CB_DATA[i] register.
-    void ProcessCBData(u32 value);
-    void ProcessCBMultiData(const u32* start_base, u32 amount);
-
     /// Returns a query's value or an empty object if the value will be deferred through a cache.
     std::optional<u64> GetQueryResult();
 

src/video_core/macro/macro.cpp

@@ -23,7 +23,7 @@
 #include "video_core/macro/macro_jit_x64.h"
 #endif
 
-MICROPROFILE_DEFINE(MacroHLE, "GPU", "Execute macro hle", MP_RGB(128, 192, 192));
+MICROPROFILE_DEFINE(MacroHLE, "GPU", "Execute macro HLE", MP_RGB(128, 192, 192));
 
 namespace Tegra {
 
@@ -107,14 +107,15 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) {
             }
         }
 
-        if (auto hle_program = hle_macros->GetHLEProgram(cache_info.hash)) {
+        auto hle_program = hle_macros->GetHLEProgram(cache_info.hash);
+        if (!hle_program || Settings::values.disable_macro_hle) {
+            maxwell3d.RefreshParameters();
+            cache_info.lle_program->Execute(parameters, method);
+        } else {
             cache_info.has_hle_program = true;
             cache_info.hle_program = std::move(hle_program);
             MICROPROFILE_SCOPE(MacroHLE);
             cache_info.hle_program->Execute(parameters, method);
-        } else {
-            maxwell3d.RefreshParameters();
-            cache_info.lle_program->Execute(parameters, method);
         }
     }
 }

src/video_core/macro/macro_hle.cpp

@@ -1,5 +1,5 @@
-// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
+// SPDX-License-Identifier: GPL-3.0-or-later
 
 #include <array>
 #include <vector>
@@ -15,28 +15,28 @@
 
 namespace Tegra {
 
-using Maxwell = Engines::Maxwell3D;
+using Maxwell3D = Engines::Maxwell3D;
 
 namespace {
 
-bool IsTopologySafe(Maxwell::Regs::PrimitiveTopology topology) {
+bool IsTopologySafe(Maxwell3D::Regs::PrimitiveTopology topology) {
     switch (topology) {
-    case Maxwell::Regs::PrimitiveTopology::Points:
+    case Maxwell3D::Regs::PrimitiveTopology::Points:
-    case Maxwell::Regs::PrimitiveTopology::Lines:
+    case Maxwell3D::Regs::PrimitiveTopology::Lines:
-    case Maxwell::Regs::PrimitiveTopology::LineLoop:
+    case Maxwell3D::Regs::PrimitiveTopology::LineLoop:
-    case Maxwell::Regs::PrimitiveTopology::LineStrip:
+    case Maxwell3D::Regs::PrimitiveTopology::LineStrip:
-    case Maxwell::Regs::PrimitiveTopology::Triangles:
+    case Maxwell3D::Regs::PrimitiveTopology::Triangles:
-    case Maxwell::Regs::PrimitiveTopology::TriangleStrip:
+    case Maxwell3D::Regs::PrimitiveTopology::TriangleStrip:
-    case Maxwell::Regs::PrimitiveTopology::TriangleFan:
+    case Maxwell3D::Regs::PrimitiveTopology::TriangleFan:
-    case Maxwell::Regs::PrimitiveTopology::LinesAdjacency:
+    case Maxwell3D::Regs::PrimitiveTopology::LinesAdjacency:
-    case Maxwell::Regs::PrimitiveTopology::LineStripAdjacency:
+    case Maxwell3D::Regs::PrimitiveTopology::LineStripAdjacency:
-    case Maxwell::Regs::PrimitiveTopology::TrianglesAdjacency:
+    case Maxwell3D::Regs::PrimitiveTopology::TrianglesAdjacency:
-    case Maxwell::Regs::PrimitiveTopology::TriangleStripAdjacency:
+    case Maxwell3D::Regs::PrimitiveTopology::TriangleStripAdjacency:
-    case Maxwell::Regs::PrimitiveTopology::Patches:
+    case Maxwell3D::Regs::PrimitiveTopology::Patches:
         return true;
-    case Maxwell::Regs::PrimitiveTopology::Quads:
+    case Maxwell3D::Regs::PrimitiveTopology::Quads:
-    case Maxwell::Regs::PrimitiveTopology::QuadStrip:
+    case Maxwell3D::Regs::PrimitiveTopology::QuadStrip:
-    case Maxwell::Regs::PrimitiveTopology::Polygon:
+    case Maxwell3D::Regs::PrimitiveTopology::Polygon:
     default:
         return false;
     }
@@ -44,20 +44,55 @@ bool IsTopologySafe(Maxwell::Regs::PrimitiveTopology topology) {
 
 class HLEMacroImpl : public CachedMacro {
 public:
-    explicit HLEMacroImpl(Engines::Maxwell3D& maxwell3d_) : maxwell3d{maxwell3d_} {}
+    explicit HLEMacroImpl(Maxwell3D& maxwell3d_) : maxwell3d{maxwell3d_} {}
 
 protected:
-    Engines::Maxwell3D& maxwell3d;
+    Maxwell3D& maxwell3d;
 };
 
+class HLE_DrawArrays final : public HLEMacroImpl {
+public:
+    explicit HLE_DrawArrays(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
+
+    void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
+        maxwell3d.RefreshParameters();
+
+        auto topology = static_cast<Maxwell3D::Regs::PrimitiveTopology>(parameters[0]);
+        maxwell3d.draw_manager->DrawArray(topology, parameters[1], parameters[2],
+                                          maxwell3d.regs.global_base_instance_index, 1);
+    }
+};
+
+class HLE_DrawIndexed final : public HLEMacroImpl {
+public:
+    explicit HLE_DrawIndexed(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
+
+    void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
+        maxwell3d.RefreshParameters();
+        maxwell3d.regs.index_buffer.start_addr_high = parameters[1];
+        maxwell3d.regs.index_buffer.start_addr_low = parameters[2];
+        maxwell3d.regs.index_buffer.format =
+            static_cast<Engines::Maxwell3D::Regs::IndexFormat>(parameters[3]);
+        maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+
+        auto topology = static_cast<Maxwell3D::Regs::PrimitiveTopology>(parameters[0]);
+        maxwell3d.draw_manager->DrawIndex(topology, 0, parameters[4],
+                                          maxwell3d.regs.global_base_vertex_index,
+                                          maxwell3d.regs.global_base_instance_index, 1);
+    }
+};
+
+/*
+ * @note: these macros have two versions, a normal and extended version, with the extended version
+ * also assigning the base vertex/instance.
+ */
 template <bool extended>
 class HLE_DrawArraysIndirect final : public HLEMacroImpl {
 public:
-    explicit HLE_DrawArraysIndirect(Engines::Maxwell3D& maxwell3d_)
+    explicit HLE_DrawArraysIndirect(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
-        : HLEMacroImpl(maxwell3d_) {}
 
     void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
-        auto topology = static_cast<Maxwell::Regs::PrimitiveTopology>(parameters[0]);
+        auto topology = static_cast<Maxwell3D::Regs::PrimitiveTopology>(parameters[0]);
         if (!maxwell3d.AnyParametersDirty() || !IsTopologySafe(topology)) {
             Fallback(parameters);
             return;
@@ -73,15 +108,15 @@ public:
         params.stride = 0;
 
         if constexpr (extended) {
-            maxwell3d.engine_state = Maxwell::EngineHint::OnHLEMacro;
+            maxwell3d.engine_state = Maxwell3D::EngineHint::OnHLEMacro;
             maxwell3d.SetHLEReplacementAttributeType(
-                0, 0x640, Maxwell::HLEReplacementAttributeType::BaseInstance);
+                0, 0x640, Maxwell3D::HLEReplacementAttributeType::BaseInstance);
         }
 
         maxwell3d.draw_manager->DrawArrayIndirect(topology);
 
         if constexpr (extended) {
-            maxwell3d.engine_state = Maxwell::EngineHint::None;
+            maxwell3d.engine_state = Maxwell3D::EngineHint::None;
             maxwell3d.replace_table.clear();
         }
     }
@@ -90,14 +125,14 @@ private:
     void Fallback(const std::vector<u32>& parameters) {
         SCOPE_EXIT({
             if (extended) {
-                maxwell3d.engine_state = Maxwell::EngineHint::None;
+                maxwell3d.engine_state = Maxwell3D::EngineHint::None;
                 maxwell3d.replace_table.clear();
             }
         });
         maxwell3d.RefreshParameters();
         const u32 instance_count = (maxwell3d.GetRegisterValue(0xD1B) & parameters[2]);
 
-        auto topology = static_cast<Maxwell::Regs::PrimitiveTopology>(parameters[0]);
+        auto topology = static_cast<Maxwell3D::Regs::PrimitiveTopology>(parameters[0]);
         const u32 vertex_first = parameters[3];
         const u32 vertex_count = parameters[1];
 
@@ -111,9 +146,9 @@ private:
         const u32 base_instance = parameters[4];
         if constexpr (extended) {
             maxwell3d.regs.global_base_instance_index = base_instance;
-            maxwell3d.engine_state = Maxwell::EngineHint::OnHLEMacro;
+            maxwell3d.engine_state = Maxwell3D::EngineHint::OnHLEMacro;
             maxwell3d.SetHLEReplacementAttributeType(
-                0, 0x640, Maxwell::HLEReplacementAttributeType::BaseInstance);
+                0, 0x640, Maxwell3D::HLEReplacementAttributeType::BaseInstance);
         }
 
         maxwell3d.draw_manager->DrawArray(topology, vertex_first, vertex_count, base_instance,
@@ -121,19 +156,23 @@ private:
 
         if constexpr (extended) {
             maxwell3d.regs.global_base_instance_index = 0;
-            maxwell3d.engine_state = Maxwell::EngineHint::None;
+            maxwell3d.engine_state = Maxwell3D::EngineHint::None;
             maxwell3d.replace_table.clear();
         }
     }
 };
 
+/*
+ * @note: these macros have two versions, a normal and extended version, with the extended version
+ * also assigning the base vertex/instance.
+ */
 template <bool extended>
 class HLE_DrawIndexedIndirect final : public HLEMacroImpl {
 public:
-    explicit HLE_DrawIndexedIndirect(Engines::Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
+    explicit HLE_DrawIndexedIndirect(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
 
     void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
-        auto topology = static_cast<Maxwell::Regs::PrimitiveTopology>(parameters[0]);
+        auto topology = static_cast<Maxwell3D::Regs::PrimitiveTopology>(parameters[0]);
         if (!maxwell3d.AnyParametersDirty() || !IsTopologySafe(topology)) {
             Fallback(parameters);
             return;
@@ -147,11 +186,11 @@ public:
         maxwell3d.regs.global_base_instance_index = base_instance;
         maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
         if constexpr (extended) {
-            maxwell3d.engine_state = Maxwell::EngineHint::OnHLEMacro;
+            maxwell3d.engine_state = Maxwell3D::EngineHint::OnHLEMacro;
             maxwell3d.SetHLEReplacementAttributeType(
-                0, 0x640, Maxwell::HLEReplacementAttributeType::BaseVertex);
+                0, 0x640, Maxwell3D::HLEReplacementAttributeType::BaseVertex);
             maxwell3d.SetHLEReplacementAttributeType(
-                0, 0x644, Maxwell::HLEReplacementAttributeType::BaseInstance);
+                0, 0x644, Maxwell3D::HLEReplacementAttributeType::BaseInstance);
         }
         auto& params = maxwell3d.draw_manager->GetIndirectParams();
         params.is_indexed = true;
@@ -167,7 +206,7 @@ public:
         maxwell3d.regs.global_base_vertex_index = 0x0;
         maxwell3d.regs.global_base_instance_index = 0x0;
         if constexpr (extended) {
-            maxwell3d.engine_state = Maxwell::EngineHint::None;
+            maxwell3d.engine_state = Maxwell3D::EngineHint::None;
             maxwell3d.replace_table.clear();
         }
     }
@@ -183,22 +222,22 @@ private:
         maxwell3d.regs.global_base_instance_index = base_instance;
         maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
         if constexpr (extended) {
-            maxwell3d.engine_state = Maxwell::EngineHint::OnHLEMacro;
+            maxwell3d.engine_state = Maxwell3D::EngineHint::OnHLEMacro;
             maxwell3d.SetHLEReplacementAttributeType(
-                0, 0x640, Maxwell::HLEReplacementAttributeType::BaseVertex);
+                0, 0x640, Maxwell3D::HLEReplacementAttributeType::BaseVertex);
             maxwell3d.SetHLEReplacementAttributeType(
-                0, 0x644, Maxwell::HLEReplacementAttributeType::BaseInstance);
+                0, 0x644, Maxwell3D::HLEReplacementAttributeType::BaseInstance);
         }
 
         maxwell3d.draw_manager->DrawIndex(
-            static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]),
+            static_cast<Tegra::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]), parameters[3],
-            parameters[3], parameters[1], element_base, base_instance, instance_count);
+            parameters[1], element_base, base_instance, instance_count);
 
         maxwell3d.regs.vertex_id_base = 0x0;
         maxwell3d.regs.global_base_vertex_index = 0x0;
         maxwell3d.regs.global_base_instance_index = 0x0;
         if constexpr (extended) {
-            maxwell3d.engine_state = Maxwell::EngineHint::None;
+            maxwell3d.engine_state = Maxwell3D::EngineHint::None;
             maxwell3d.replace_table.clear();
         }
     }
@@ -206,13 +245,13 @@ private:
 
 class HLE_MultiLayerClear final : public HLEMacroImpl {
 public:
-    explicit HLE_MultiLayerClear(Engines::Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
+    explicit HLE_MultiLayerClear(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
 
     void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
         maxwell3d.RefreshParameters();
         ASSERT(parameters.size() == 1);
 
-        const Engines::Maxwell3D::Regs::ClearSurface clear_params{parameters[0]};
+        const Maxwell3D::Regs::ClearSurface clear_params{parameters[0]};
         const u32 rt_index = clear_params.RT;
         const u32 num_layers = maxwell3d.regs.rt[rt_index].depth;
         ASSERT(clear_params.layer == 0);
@@ -224,11 +263,10 @@ public:
 
 class HLE_MultiDrawIndexedIndirectCount final : public HLEMacroImpl {
 public:
-    explicit HLE_MultiDrawIndexedIndirectCount(Engines::Maxwell3D& maxwell3d_)
+    explicit HLE_MultiDrawIndexedIndirectCount(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
-        : HLEMacroImpl(maxwell3d_) {}
 
     void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
-        const auto topology = static_cast<Maxwell::Regs::PrimitiveTopology>(parameters[2]);
+        const auto topology = static_cast<Maxwell3D::Regs::PrimitiveTopology>(parameters[2]);
         if (!IsTopologySafe(topology)) {
             Fallback(parameters);
             return;
@@ -258,23 +296,24 @@ public:
         params.max_draw_counts = draw_count;
         params.stride = stride;
         maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
-        maxwell3d.engine_state = Maxwell::EngineHint::OnHLEMacro;
+        maxwell3d.engine_state = Maxwell3D::EngineHint::OnHLEMacro;
-        maxwell3d.SetHLEReplacementAttributeType(0, 0x640,
-                                                 Maxwell::HLEReplacementAttributeType::BaseVertex);
         maxwell3d.SetHLEReplacementAttributeType(
-            0, 0x644, Maxwell::HLEReplacementAttributeType::BaseInstance);
+            0, 0x640, Maxwell3D::HLEReplacementAttributeType::BaseVertex);
+        maxwell3d.SetHLEReplacementAttributeType(
+            0, 0x644, Maxwell3D::HLEReplacementAttributeType::BaseInstance);
+        maxwell3d.SetHLEReplacementAttributeType(0, 0x648,
+                                                 Maxwell3D::HLEReplacementAttributeType::DrawID);
         maxwell3d.draw_manager->DrawIndexedIndirect(topology, 0, estimate);
-        maxwell3d.engine_state = Maxwell::EngineHint::None;
+        maxwell3d.engine_state = Maxwell3D::EngineHint::None;
         maxwell3d.replace_table.clear();
     }
 
 private:
     void Fallback(const std::vector<u32>& parameters) {
         SCOPE_EXIT({
-            // Clean everything.
             // Clean everything.
             maxwell3d.regs.vertex_id_base = 0x0;
-            maxwell3d.engine_state = Maxwell::EngineHint::None;
+            maxwell3d.engine_state = Maxwell3D::EngineHint::None;
             maxwell3d.replace_table.clear();
         });
         maxwell3d.RefreshParameters();
@@ -284,7 +323,7 @@ private:
             // Nothing to do.
             return;
         }
-        const auto topology = static_cast<Maxwell::Regs::PrimitiveTopology>(parameters[2]);
+        const auto topology = static_cast<Maxwell3D::Regs::PrimitiveTopology>(parameters[2]);
         const u32 padding = parameters[3];
         const std::size_t max_draws = parameters[4];
 
@@ -298,11 +337,13 @@ private:
             const u32 base_vertex = parameters[base + 3];
             const u32 base_instance = parameters[base + 4];
             maxwell3d.regs.vertex_id_base = base_vertex;
-            maxwell3d.engine_state = Maxwell::EngineHint::OnHLEMacro;
+            maxwell3d.engine_state = Maxwell3D::EngineHint::OnHLEMacro;
             maxwell3d.SetHLEReplacementAttributeType(
-                0, 0x640, Maxwell::HLEReplacementAttributeType::BaseVertex);
+                0, 0x640, Maxwell3D::HLEReplacementAttributeType::BaseVertex);
             maxwell3d.SetHLEReplacementAttributeType(
-                0, 0x644, Maxwell::HLEReplacementAttributeType::BaseInstance);
+                0, 0x644, Maxwell3D::HLEReplacementAttributeType::BaseInstance);
+            maxwell3d.CallMethod(0x8e3, 0x648, true);
+            maxwell3d.CallMethod(0x8e4, static_cast<u32>(index), true);
             maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
             maxwell3d.draw_manager->DrawIndex(topology, parameters[base + 2], parameters[base],
                                               base_vertex, base_instance, parameters[base + 1]);
@@ -312,7 +353,7 @@ private:
 
 class HLE_C713C83D8F63CCF3 final : public HLEMacroImpl {
 public:
-    explicit HLE_C713C83D8F63CCF3(Engines::Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
+    explicit HLE_C713C83D8F63CCF3(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
 
     void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
         maxwell3d.RefreshParameters();
@@ -328,7 +369,7 @@ public:
 
 class HLE_D7333D26E0A93EDE final : public HLEMacroImpl {
 public:
-    explicit HLE_D7333D26E0A93EDE(Engines::Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
+    explicit HLE_D7333D26E0A93EDE(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
 
     void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
         maxwell3d.RefreshParameters();
@@ -344,7 +385,7 @@ public:
 
 class HLE_BindShader final : public HLEMacroImpl {
 public:
-    explicit HLE_BindShader(Engines::Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
+    explicit HLE_BindShader(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
 
     void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
         maxwell3d.RefreshParameters();
@@ -374,7 +415,7 @@ public:
 
 class HLE_SetRasterBoundingBox final : public HLEMacroImpl {
 public:
-    explicit HLE_SetRasterBoundingBox(Engines::Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
+    explicit HLE_SetRasterBoundingBox(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
 
     void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
         maxwell3d.RefreshParameters();
@@ -387,60 +428,156 @@ public:
     }
 };
 
+template <size_t base_size>
+class HLE_ClearConstBuffer final : public HLEMacroImpl {
+public:
+    explicit HLE_ClearConstBuffer(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
+
+    void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
+        maxwell3d.RefreshParameters();
+        static constexpr std::array<u32, base_size> zeroes{};
+        auto& regs = maxwell3d.regs;
+        regs.const_buffer.size = static_cast<u32>(base_size);
+        regs.const_buffer.address_high = parameters[0];
+        regs.const_buffer.address_low = parameters[1];
+        regs.const_buffer.offset = 0;
+        maxwell3d.ProcessCBMultiData(zeroes.data(), parameters[2] * 4);
+    }
+};
+
+class HLE_ClearMemory final : public HLEMacroImpl {
+public:
+    explicit HLE_ClearMemory(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
+
+    void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
+        maxwell3d.RefreshParameters();
+
+        const u32 needed_memory = parameters[2] / sizeof(u32);
+        if (needed_memory > zero_memory.size()) {
+            zero_memory.resize(needed_memory, 0);
+        }
+        auto& regs = maxwell3d.regs;
+        regs.upload.line_length_in = parameters[2];
+        regs.upload.line_count = 1;
+        regs.upload.dest.address_high = parameters[0];
+        regs.upload.dest.address_low = parameters[1];
+        maxwell3d.CallMethod(static_cast<size_t>(MAXWELL3D_REG_INDEX(launch_dma)), 0x1011, true);
+        maxwell3d.CallMultiMethod(static_cast<size_t>(MAXWELL3D_REG_INDEX(inline_data)),
+                                  zero_memory.data(), needed_memory, needed_memory);
+    }
+
+private:
+    std::vector<u32> zero_memory;
+};
+
+class HLE_TransformFeedbackSetup final : public HLEMacroImpl {
+public:
+    explicit HLE_TransformFeedbackSetup(Maxwell3D& maxwell3d_) : HLEMacroImpl(maxwell3d_) {}
+
+    void Execute(const std::vector<u32>& parameters, [[maybe_unused]] u32 method) override {
+        maxwell3d.RefreshParameters();
+
+        auto& regs = maxwell3d.regs;
+        regs.transform_feedback_enabled = 1;
+        regs.transform_feedback.buffers[0].start_offset = 0;
+        regs.transform_feedback.buffers[1].start_offset = 0;
+        regs.transform_feedback.buffers[2].start_offset = 0;
+        regs.transform_feedback.buffers[3].start_offset = 0;
+
+        regs.upload.line_length_in = 4;
+        regs.upload.line_count = 1;
+        regs.upload.dest.address_high = parameters[0];
+        regs.upload.dest.address_low = parameters[1];
+        maxwell3d.CallMethod(static_cast<size_t>(MAXWELL3D_REG_INDEX(launch_dma)), 0x1011, true);
+        maxwell3d.CallMethod(static_cast<size_t>(MAXWELL3D_REG_INDEX(inline_data)),
+                             regs.transform_feedback.controls[0].stride, true);
+    }
+};
+
 } // Anonymous namespace
 
-HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d_) : maxwell3d{maxwell3d_} {
+HLEMacro::HLEMacro(Maxwell3D& maxwell3d_) : maxwell3d{maxwell3d_} {
+    builders.emplace(0xDD6A7FA92A7D2674ULL,
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                             return std::make_unique<HLE_DrawArrays>(maxwell3d__);
+                         }));
     builders.emplace(0x0D61FC9FAAC9FCADULL,
-                     std::function<std::unique_ptr<CachedMacro>(Engines::Maxwell3D&)>(
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
-                         [](Engines::Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
                              return std::make_unique<HLE_DrawArraysIndirect<false>>(maxwell3d__);
                          }));
     builders.emplace(0x8A4D173EB99A8603ULL,
-                     std::function<std::unique_ptr<CachedMacro>(Engines::Maxwell3D&)>(
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
-                         [](Engines::Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
                              return std::make_unique<HLE_DrawArraysIndirect<true>>(maxwell3d__);
                          }));
+    builders.emplace(0x2DB33AADB741839CULL,
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                             return std::make_unique<HLE_DrawIndexed>(maxwell3d__);
+                         }));
     builders.emplace(0x771BB18C62444DA0ULL,
-                     std::function<std::unique_ptr<CachedMacro>(Engines::Maxwell3D&)>(
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
-                         [](Engines::Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
                              return std::make_unique<HLE_DrawIndexedIndirect<false>>(maxwell3d__);
                          }));
     builders.emplace(0x0217920100488FF7ULL,
-                     std::function<std::unique_ptr<CachedMacro>(Engines::Maxwell3D&)>(
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
-                         [](Engines::Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
                              return std::make_unique<HLE_DrawIndexedIndirect<true>>(maxwell3d__);
                          }));
     builders.emplace(0x3F5E74B9C9A50164ULL,
-                     std::function<std::unique_ptr<CachedMacro>(Engines::Maxwell3D&)>(
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
-                         [](Engines::Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
                              return std::make_unique<HLE_MultiDrawIndexedIndirectCount>(
                                  maxwell3d__);
                          }));
     builders.emplace(0xEAD26C3E2109B06BULL,
-                     std::function<std::unique_ptr<CachedMacro>(Engines::Maxwell3D&)>(
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
-                         [](Engines::Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
                              return std::make_unique<HLE_MultiLayerClear>(maxwell3d__);
                          }));
     builders.emplace(0xC713C83D8F63CCF3ULL,
-                     std::function<std::unique_ptr<CachedMacro>(Engines::Maxwell3D&)>(
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
-                         [](Engines::Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
                              return std::make_unique<HLE_C713C83D8F63CCF3>(maxwell3d__);
                          }));
     builders.emplace(0xD7333D26E0A93EDEULL,
-                     std::function<std::unique_ptr<CachedMacro>(Engines::Maxwell3D&)>(
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
-                         [](Engines::Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
                              return std::make_unique<HLE_D7333D26E0A93EDE>(maxwell3d__);
                          }));
     builders.emplace(0xEB29B2A09AA06D38ULL,
-                     std::function<std::unique_ptr<CachedMacro>(Engines::Maxwell3D&)>(
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
-                         [](Engines::Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
                              return std::make_unique<HLE_BindShader>(maxwell3d__);
                          }));
     builders.emplace(0xDB1341DBEB4C8AF7ULL,
-                     std::function<std::unique_ptr<CachedMacro>(Engines::Maxwell3D&)>(
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
-                         [](Engines::Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
                              return std::make_unique<HLE_SetRasterBoundingBox>(maxwell3d__);
                          }));
+    builders.emplace(0x6C97861D891EDf7EULL,
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                             return std::make_unique<HLE_ClearConstBuffer<0x5F00>>(maxwell3d__);
+                         }));
+    builders.emplace(0xD246FDDF3A6173D7ULL,
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                             return std::make_unique<HLE_ClearConstBuffer<0x7000>>(maxwell3d__);
+                         }));
+    builders.emplace(0xEE4D0004BEC8ECF4ULL,
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                             return std::make_unique<HLE_ClearMemory>(maxwell3d__);
+                         }));
+    builders.emplace(0xFC0CF27F5FFAA661ULL,
+                     std::function<std::unique_ptr<CachedMacro>(Maxwell3D&)>(
+                         [](Maxwell3D& maxwell3d__) -> std::unique_ptr<CachedMacro> {
+                             return std::make_unique<HLE_TransformFeedbackSetup>(maxwell3d__);
+                         }));
 }
 
 HLEMacro::~HLEMacro() = default;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -357,21 +357,21 @@ void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size, VideoCommon::CacheType
     if (addr == 0 || size == 0) {
         return;
     }
-    if (bool(which & VideoCommon::CacheType::TextureCache)) {
+    if (True(which & VideoCommon::CacheType::TextureCache)) {
         std::scoped_lock lock{texture_cache.mutex};
         texture_cache.DownloadMemory(addr, size);
     }
-    if ((bool(which & VideoCommon::CacheType::BufferCache))) {
+    if ((True(which & VideoCommon::CacheType::BufferCache))) {
         std::scoped_lock lock{buffer_cache.mutex};
         buffer_cache.DownloadMemory(addr, size);
     }
-    if ((bool(which & VideoCommon::CacheType::QueryCache))) {
+    if ((True(which & VideoCommon::CacheType::QueryCache))) {
         query_cache.FlushRegion(addr, size);
     }
 }
 
 bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheType which) {
-    if ((bool(which & VideoCommon::CacheType::BufferCache))) {
+    if ((True(which & VideoCommon::CacheType::BufferCache))) {
         std::scoped_lock lock{buffer_cache.mutex};
         if (buffer_cache.IsRegionGpuModified(addr, size)) {
             return true;
@@ -380,7 +380,7 @@ bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheT
     if (!Settings::IsGPULevelHigh()) {
         return false;
     }
-    if (bool(which & VideoCommon::CacheType::TextureCache)) {
+    if (True(which & VideoCommon::CacheType::TextureCache)) {
         std::scoped_lock lock{texture_cache.mutex};
         return texture_cache.IsRegionGpuModified(addr, size);
     }
@@ -392,18 +392,18 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size, VideoCommon::Cache
     if (addr == 0 || size == 0) {
         return;
     }
-    if (bool(which & VideoCommon::CacheType::TextureCache)) {
+    if (True(which & VideoCommon::CacheType::TextureCache)) {
         std::scoped_lock lock{texture_cache.mutex};
         texture_cache.WriteMemory(addr, size);
     }
-    if (bool(which & VideoCommon::CacheType::BufferCache)) {
+    if (True(which & VideoCommon::CacheType::BufferCache)) {
         std::scoped_lock lock{buffer_cache.mutex};
         buffer_cache.WriteMemory(addr, size);
     }
-    if (bool(which & VideoCommon::CacheType::ShaderCache)) {
+    if (True(which & VideoCommon::CacheType::ShaderCache)) {
         shader_cache.InvalidateRegion(addr, size);
     }
-    if (bool(which & VideoCommon::CacheType::QueryCache)) {
+    if (True(which & VideoCommon::CacheType::QueryCache)) {
         query_cache.InvalidateRegion(addr, size);
     }
 }

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -431,21 +431,21 @@ void RasterizerVulkan::FlushRegion(VAddr addr, u64 size, VideoCommon::CacheType
     if (addr == 0 || size == 0) {
         return;
     }
-    if (bool(which & VideoCommon::CacheType::TextureCache)) {
+    if (True(which & VideoCommon::CacheType::TextureCache)) {
         std::scoped_lock lock{texture_cache.mutex};
         texture_cache.DownloadMemory(addr, size);
     }
-    if ((bool(which & VideoCommon::CacheType::BufferCache))) {
+    if ((True(which & VideoCommon::CacheType::BufferCache))) {
         std::scoped_lock lock{buffer_cache.mutex};
         buffer_cache.DownloadMemory(addr, size);
     }
-    if ((bool(which & VideoCommon::CacheType::QueryCache))) {
+    if ((True(which & VideoCommon::CacheType::QueryCache))) {
         query_cache.FlushRegion(addr, size);
     }
 }
 
 bool RasterizerVulkan::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheType which) {
-    if ((bool(which & VideoCommon::CacheType::BufferCache))) {
+    if ((True(which & VideoCommon::CacheType::BufferCache))) {
         std::scoped_lock lock{buffer_cache.mutex};
         if (buffer_cache.IsRegionGpuModified(addr, size)) {
             return true;
@@ -454,7 +454,7 @@ bool RasterizerVulkan::MustFlushRegion(VAddr addr, u64 size, VideoCommon::CacheT
     if (!Settings::IsGPULevelHigh()) {
         return false;
     }
-    if (bool(which & VideoCommon::CacheType::TextureCache)) {
+    if (True(which & VideoCommon::CacheType::TextureCache)) {
         std::scoped_lock lock{texture_cache.mutex};
         return texture_cache.IsRegionGpuModified(addr, size);
     }
@@ -465,18 +465,18 @@ void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size, VideoCommon::Cache
     if (addr == 0 || size == 0) {
         return;
     }
-    if (bool(which & VideoCommon::CacheType::TextureCache)) {
+    if (True(which & VideoCommon::CacheType::TextureCache)) {
         std::scoped_lock lock{texture_cache.mutex};
         texture_cache.WriteMemory(addr, size);
     }
-    if ((bool(which & VideoCommon::CacheType::BufferCache))) {
+    if ((True(which & VideoCommon::CacheType::BufferCache))) {
         std::scoped_lock lock{buffer_cache.mutex};
         buffer_cache.WriteMemory(addr, size);
     }
-    if ((bool(which & VideoCommon::CacheType::QueryCache))) {
+    if ((True(which & VideoCommon::CacheType::QueryCache))) {
         query_cache.InvalidateRegion(addr, size);
     }
-    if ((bool(which & VideoCommon::CacheType::ShaderCache))) {
+    if ((True(which & VideoCommon::CacheType::ShaderCache))) {
         pipeline_cache.InvalidateRegion(addr, size);
     }
 }

src/video_core/renderer_vulkan/vk_state_tracker.cpp

@@ -251,4 +251,4 @@ void StateTracker::InvalidateState() {
 StateTracker::StateTracker()
     : flags{&default_flags}, default_flags{}, invalidation_flags{MakeInvalidationFlags()} {}
 
-} // namespace Vulkan
+} // namespace Vulkan

src/video_core/shader_environment.cpp

@@ -357,6 +357,8 @@ std::optional<Shader::ReplaceConstant> GraphicsEnvironment::GetReplaceConstBuffe
             return Shader::ReplaceConstant::BaseVertex;
         case Tegra::Engines::Maxwell3D::HLEReplacementAttributeType::BaseInstance:
             return Shader::ReplaceConstant::BaseInstance;
+        case Tegra::Engines::Maxwell3D::HLEReplacementAttributeType::DrawID:
+            return Shader::ReplaceConstant::DrawID;
         default:
             UNREACHABLE();
         }

src/yuzu/configuration/config.cpp

@@ -563,6 +563,7 @@ void Config::ReadDebuggingValues() {
     ReadBasicSetting(Settings::values.reporting_services);
     ReadBasicSetting(Settings::values.quest_flag);
     ReadBasicSetting(Settings::values.disable_macro_jit);
+    ReadBasicSetting(Settings::values.disable_macro_hle);
     ReadBasicSetting(Settings::values.extended_logging);
     ReadBasicSetting(Settings::values.use_debug_asserts);
     ReadBasicSetting(Settings::values.use_auto_stub);
@@ -1200,6 +1201,7 @@ void Config::SaveDebuggingValues() {
     WriteBasicSetting(Settings::values.quest_flag);
     WriteBasicSetting(Settings::values.use_debug_asserts);
     WriteBasicSetting(Settings::values.disable_macro_jit);
+    WriteBasicSetting(Settings::values.disable_macro_hle);
     WriteBasicSetting(Settings::values.enable_all_controllers);
     WriteBasicSetting(Settings::values.create_crash_dumps);
     WriteBasicSetting(Settings::values.perform_vulkan_check);

src/yuzu/configuration/configure_debug.cpp

@@ -73,6 +73,8 @@ void ConfigureDebug::SetConfiguration() {
     ui->dump_macros->setChecked(Settings::values.dump_macros.GetValue());
     ui->disable_macro_jit->setEnabled(runtime_lock);
     ui->disable_macro_jit->setChecked(Settings::values.disable_macro_jit.GetValue());
+    ui->disable_macro_hle->setEnabled(runtime_lock);
+    ui->disable_macro_hle->setChecked(Settings::values.disable_macro_hle.GetValue());
     ui->disable_loop_safety_checks->setEnabled(runtime_lock);
     ui->disable_loop_safety_checks->setChecked(
         Settings::values.disable_shader_loop_safety_checks.GetValue());
@@ -117,6 +119,7 @@ void ConfigureDebug::ApplyConfiguration() {
     Settings::values.disable_shader_loop_safety_checks =
         ui->disable_loop_safety_checks->isChecked();
     Settings::values.disable_macro_jit = ui->disable_macro_jit->isChecked();
+    Settings::values.disable_macro_hle = ui->disable_macro_hle->isChecked();
     Settings::values.extended_logging = ui->extended_logging->isChecked();
     Settings::values.perform_vulkan_check = ui->perform_vulkan_check->isChecked();
     UISettings::values.disable_web_applet = ui->disable_web_applet->isChecked();

src/yuzu/configuration/configure_debug.ui

@@ -176,7 +176,7 @@
         </property>
        </widget>
       </item>
-      <item row="0" column="2">
+      <item row="1" column="2">
        <widget class="QCheckBox" name="dump_macros">
         <property name="enabled">
          <bool>true</bool>
@@ -202,6 +202,19 @@
         </property>
        </widget>
       </item>
+      <item row="0" column="2">
+       <widget class="QCheckBox" name="disable_macro_hle">
+        <property name="enabled">
+         <bool>true</bool>
+        </property>
+        <property name="toolTip">
+         <string>When checked, it disables the macro HLE functions. Enabling this makes games run slower</string>
+        </property>
+        <property name="text">
+         <string>Disable Macro HLE</string>
+        </property>
+       </widget>
+      </item>
       <item row="1" column="0">
        <widget class="QCheckBox" name="enable_shader_feedback">
         <property name="toolTip">

src/yuzu_cmd/config.cpp

@@ -348,6 +348,7 @@ void Config::ReadValues() {
     ReadSetting("Debugging", Settings::values.use_debug_asserts);
     ReadSetting("Debugging", Settings::values.use_auto_stub);
     ReadSetting("Debugging", Settings::values.disable_macro_jit);
+    ReadSetting("Debugging", Settings::values.disable_macro_hle);
     ReadSetting("Debugging", Settings::values.use_gdbstub);
     ReadSetting("Debugging", Settings::values.gdbstub_port);