video_core: Implement maxwell3d draw manager and split draw logic

src/video_core/CMakeLists.txt

@@ -33,6 +33,8 @@ add_library(video_core STATIC
     engines/sw_blitter/converter.cpp
     engines/sw_blitter/converter.h
     engines/const_buffer_info.h
+    engines/draw_manager.cpp
+    engines/draw_manager.h
     engines/engine_interface.h
     engines/engine_upload.cpp
     engines/engine_upload.h

src/video_core/buffer_cache/buffer_cache.h

@@ -26,6 +26,7 @@
 #include "video_core/control/channel_state_cache.h"
 #include "video_core/delayed_destruction_ring.h"
 #include "video_core/dirty_flags.h"
+#include "video_core/engines/draw_manager.h"
 #include "video_core/engines/kepler_compute.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/memory_manager.h"
@@ -664,9 +665,10 @@ void BufferCache<P>::BindHostGeometryBuffers(bool is_indexed) {
     if (is_indexed) {
         BindHostIndexBuffer();
     } else if constexpr (!HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) {
-        const auto& regs = maxwell3d->regs;
+        const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
-        if (regs.draw.topology == Maxwell::PrimitiveTopology::Quads) {
+        if (draw_state.topology == Maxwell::PrimitiveTopology::Quads) {
-            runtime.BindQuadArrayIndexBuffer(regs.vertex_buffer.first, regs.vertex_buffer.count);
+            runtime.BindQuadArrayIndexBuffer(draw_state.vertex_buffer.first,
+                                             draw_state.vertex_buffer.count);
         }
     }
     BindHostVertexBuffers();
@@ -993,28 +995,29 @@ void BufferCache<P>::BindHostIndexBuffer() {
     TouchBuffer(buffer, index_buffer.buffer_id);
     const u32 offset = buffer.Offset(index_buffer.cpu_addr);
     const u32 size = index_buffer.size;
-    if (maxwell3d->inline_index_draw_indexes.size()) {
+    const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
+    if (!draw_state.inline_index_draw_indexes.empty()) {
         if constexpr (USE_MEMORY_MAPS) {
             auto upload_staging = runtime.UploadStagingBuffer(size);
             std::array<BufferCopy, 1> copies{
                 {BufferCopy{.src_offset = upload_staging.offset, .dst_offset = 0, .size = size}}};
             std::memcpy(upload_staging.mapped_span.data(),
-                        maxwell3d->inline_index_draw_indexes.data(), size);
+                        draw_state.inline_index_draw_indexes.data(), size);
             runtime.CopyBuffer(buffer, upload_staging.buffer, copies);
         } else {
-            buffer.ImmediateUpload(0, maxwell3d->inline_index_draw_indexes);
+            buffer.ImmediateUpload(0, draw_state.inline_index_draw_indexes);
         }
     } else {
         SynchronizeBuffer(buffer, index_buffer.cpu_addr, size);
     }
     if constexpr (HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) {
-        const u32 new_offset = offset + maxwell3d->regs.index_buffer.first *
+        const u32 new_offset =
-                                            maxwell3d->regs.index_buffer.FormatSizeInBytes();
+            offset + draw_state.index_buffer.first * draw_state.index_buffer.FormatSizeInBytes();
         runtime.BindIndexBuffer(buffer, new_offset, size);
     } else {
-        runtime.BindIndexBuffer(maxwell3d->regs.draw.topology, maxwell3d->regs.index_buffer.format,
+        runtime.BindIndexBuffer(draw_state.topology, draw_state.index_buffer.format,
-                                maxwell3d->regs.index_buffer.first,
+                                draw_state.index_buffer.first, draw_state.index_buffer.count,
-                                maxwell3d->regs.index_buffer.count, buffer, offset, size);
+                                buffer, offset, size);
     }
 }
 
@@ -1282,15 +1285,16 @@ template <class P>
 void BufferCache<P>::UpdateIndexBuffer() {
     // We have to check for the dirty flags and index count
     // The index count is currently changed without updating the dirty flags
-    const auto& index_array = maxwell3d->regs.index_buffer;
+    const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
+    const auto& index_array = draw_state.index_buffer;
     auto& flags = maxwell3d->dirty.flags;
     if (!flags[Dirty::IndexBuffer] && last_index_count == index_array.count) {
         return;
     }
     flags[Dirty::IndexBuffer] = false;
     last_index_count = index_array.count;
-    if (maxwell3d->inline_index_draw_indexes.size()) {
+    if (!draw_state.inline_index_draw_indexes.empty()) {
-        auto inline_index_size = static_cast<u32>(maxwell3d->inline_index_draw_indexes.size());
+        auto inline_index_size = static_cast<u32>(draw_state.inline_index_draw_indexes.size());
         index_buffer = Binding{
             .cpu_addr = 0,
             .size = inline_index_size,

src/video_core/engines/draw_manager.cpp

@@ -0,0 +1,191 @@
+// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "video_core/dirty_flags.h"
+#include "video_core/engines/draw_manager.h"
+#include "video_core/rasterizer_interface.h"
+
+namespace Tegra::Engines {
+DrawManager::DrawManager(Maxwell3D* maxwell3d_) : maxwell3d(maxwell3d_) {}
+
+void DrawManager::ProcessMethodCall(u32 method, u32 argument) {
+    const auto& regs{maxwell3d->regs};
+    switch (method) {
+    case MAXWELL3D_REG_INDEX(clear_surface):
+        return Clear(1);
+    case MAXWELL3D_REG_INDEX(draw.begin):
+        return DrawBegin();
+    case MAXWELL3D_REG_INDEX(draw.end):
+        return DrawEnd();
+    case MAXWELL3D_REG_INDEX(vertex_buffer.first):
+    case MAXWELL3D_REG_INDEX(vertex_buffer.count):
+    case MAXWELL3D_REG_INDEX(index_buffer.first):
+        break;
+    case MAXWELL3D_REG_INDEX(index_buffer.count):
+        draw_state.draw_indexed = true;
+        break;
+    case MAXWELL3D_REG_INDEX(index_buffer32_subsequent):
+    case MAXWELL3D_REG_INDEX(index_buffer16_subsequent):
+    case MAXWELL3D_REG_INDEX(index_buffer8_subsequent):
+        draw_state.instance_count++;
+        [[fallthrough]];
+    case MAXWELL3D_REG_INDEX(index_buffer32_first):
+    case MAXWELL3D_REG_INDEX(index_buffer16_first):
+    case MAXWELL3D_REG_INDEX(index_buffer8_first):
+        return DrawIndexSmall(argument);
+    case MAXWELL3D_REG_INDEX(draw_inline_index):
+        SetInlineIndexBuffer(argument);
+        break;
+    case MAXWELL3D_REG_INDEX(inline_index_2x16.even):
+        SetInlineIndexBuffer(regs.inline_index_2x16.even);
+        SetInlineIndexBuffer(regs.inline_index_2x16.odd);
+        break;
+    case MAXWELL3D_REG_INDEX(inline_index_4x8.index0):
+        SetInlineIndexBuffer(regs.inline_index_4x8.index0);
+        SetInlineIndexBuffer(regs.inline_index_4x8.index1);
+        SetInlineIndexBuffer(regs.inline_index_4x8.index2);
+        SetInlineIndexBuffer(regs.inline_index_4x8.index3);
+        break;
+    case MAXWELL3D_REG_INDEX(topology_override):
+        use_topology_override = true;
+        break;
+    default:
+        break;
+    }
+}
+
+void DrawManager::Clear(u32 layer_count) {
+    maxwell3d->rasterizer->Clear(layer_count);
+}
+
+void DrawManager::DrawDeferred() {
+    if (draw_state.draw_mode != DrawMode::Instance || draw_state.instance_count == 0)
+        return;
+    DrawEnd(draw_state.instance_count + 1, true);
+    draw_state.instance_count = 0;
+}
+
+void DrawManager::DrawArray(PrimitiveTopology topology, u32 vertex_first, u32 vertex_count,
+                            u32 base_instance, u32 num_instances) {
+    draw_state.topology = topology;
+    draw_state.vertex_buffer.first = vertex_first;
+    draw_state.vertex_buffer.count = vertex_count;
+    draw_state.base_instance = base_instance;
+    ProcessDraw(false, num_instances);
+}
+
+void DrawManager::DrawIndex(PrimitiveTopology topology, u32 index_first, u32 index_count,
+                            u32 base_index, u32 base_instance, u32 num_instances) {
+    const auto& regs{maxwell3d->regs};
+    draw_state.topology = topology;
+    draw_state.index_buffer = regs.index_buffer;
+    draw_state.index_buffer.first = index_first;
+    draw_state.index_buffer.count = index_count;
+    draw_state.base_index = base_index;
+    draw_state.base_instance = base_instance;
+    ProcessDraw(true, num_instances);
+}
+
+void DrawManager::SetInlineIndexBuffer(u32 index) {
+    draw_state.inline_index_draw_indexes.push_back(static_cast<u8>(index & 0x000000ff));
+    draw_state.inline_index_draw_indexes.push_back(static_cast<u8>((index & 0x0000ff00) >> 8));
+    draw_state.inline_index_draw_indexes.push_back(static_cast<u8>((index & 0x00ff0000) >> 16));
+    draw_state.inline_index_draw_indexes.push_back(static_cast<u8>((index & 0xff000000) >> 24));
+    draw_state.draw_mode = DrawMode::InlineIndex;
+}
+
+void DrawManager::DrawBegin() {
+    const auto& regs{maxwell3d->regs};
+    auto reset_instance_count = regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::First;
+    auto increment_instance_count =
+        regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Subsequent;
+    if (reset_instance_count) {
+        DrawDeferred();
+        draw_state.instance_count = 0;
+        draw_state.draw_mode = DrawMode::General;
+    } else if (increment_instance_count) {
+        draw_state.instance_count++;
+        draw_state.draw_mode = DrawMode::Instance;
+    }
+
+    draw_state.topology = regs.draw.topology;
+}
+
+void DrawManager::DrawEnd(u32 instance_count, bool force_draw) {
+    const auto& regs{maxwell3d->regs};
+    switch (draw_state.draw_mode) {
+    case DrawMode::Instance:
+        if (!force_draw)
+            break;
+        [[fallthrough]];
+    case DrawMode::General:
+        draw_state.base_instance = regs.global_base_instance_index;
+        draw_state.base_index = regs.global_base_vertex_index;
+        if (draw_state.draw_indexed) {
+            draw_state.index_buffer = regs.index_buffer;
+            ProcessDraw(true, instance_count);
+        } else {
+            draw_state.vertex_buffer = regs.vertex_buffer;
+            ProcessDraw(false, instance_count);
+        }
+        draw_state.draw_indexed = false;
+        break;
+    case DrawMode::InlineIndex:
+        draw_state.base_instance = regs.global_base_instance_index;
+        draw_state.base_index = regs.global_base_vertex_index;
+        draw_state.index_buffer = regs.index_buffer;
+        draw_state.index_buffer.count =
+            static_cast<u32>(draw_state.inline_index_draw_indexes.size() / 4);
+        draw_state.index_buffer.format = Maxwell3D::Regs::IndexFormat::UnsignedInt;
+        ProcessDraw(true, instance_count);
+        draw_state.inline_index_draw_indexes.clear();
+        break;
+    }
+}
+
+void DrawManager::DrawIndexSmall(u32 argument) {
+    const auto& regs{maxwell3d->regs};
+    IndexBufferSmall index_small_params{argument};
+    draw_state.base_instance = regs.global_base_instance_index;
+    draw_state.base_index = regs.global_base_vertex_index;
+    draw_state.index_buffer = regs.index_buffer;
+    draw_state.index_buffer.first = index_small_params.first;
+    draw_state.index_buffer.count = index_small_params.count;
+    draw_state.topology = index_small_params.topology;
+    maxwell3d->dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+    ProcessDraw(true, 1);
+}
+
+void DrawManager::ProcessTopologyOverride() {
+    if (!use_topology_override)
+        return;
+
+    const auto& regs{maxwell3d->regs};
+    switch (regs.topology_override) {
+    case PrimitiveTopologyOverride::None:
+        break;
+    case PrimitiveTopologyOverride::Points:
+        draw_state.topology = PrimitiveTopology::Points;
+        break;
+    case PrimitiveTopologyOverride::Lines:
+        draw_state.topology = PrimitiveTopology::Lines;
+        break;
+    case PrimitiveTopologyOverride::LineStrip:
+        draw_state.topology = PrimitiveTopology::LineStrip;
+        break;
+    default:
+        draw_state.topology = static_cast<PrimitiveTopology>(regs.topology_override);
+        break;
+    }
+}
+
+void DrawManager::ProcessDraw(bool draw_indexed, u32 instance_count) {
+    LOG_TRACE(HW_GPU, "called, topology={}, count={}", draw_state.topology.Value(),
+              draw_indexed ? draw_state.index_buffer.count : draw_state.vertex_buffer.count);
+
+    ProcessTopologyOverride();
+
+    if (maxwell3d->ShouldExecute())
+        maxwell3d->rasterizer->Draw(draw_indexed, instance_count);
+}
+} // namespace Tegra::Engines

src/video_core/engines/draw_manager.h

@@ -0,0 +1,69 @@
+// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+#include "common/common_types.h"
+#include "video_core/engines/maxwell_3d.h"
+
+namespace VideoCore {
+class RasterizerInterface;
+}
+
+namespace Tegra::Engines {
+using PrimitiveTopology = Maxwell3D::Regs::PrimitiveTopology;
+using PrimitiveTopologyOverride = Maxwell3D::Regs::PrimitiveTopologyOverride;
+using IndexBuffer = Maxwell3D::Regs::IndexBuffer;
+using VertexBuffer = Maxwell3D::Regs::VertexBuffer;
+using IndexBufferSmall = Maxwell3D::Regs::IndexBufferSmall;
+
+class DrawManager {
+public:
+    enum class DrawMode : u32 { General = 0, Instance, InlineIndex };
+    struct State {
+        PrimitiveTopology topology{};
+        DrawMode draw_mode{};
+        bool draw_indexed{};
+        u32 base_index{};
+        VertexBuffer vertex_buffer;
+        IndexBuffer index_buffer;
+        u32 base_instance{};
+        u32 instance_count{};
+        std::vector<u8> inline_index_draw_indexes;
+    };
+
+    explicit DrawManager(Maxwell3D* maxwell_3d);
+
+    void ProcessMethodCall(u32 method, u32 argument);
+
+    void Clear(u32 layer_count);
+
+    void DrawDeferred();
+
+    void DrawArray(PrimitiveTopology topology, u32 vertex_first, u32 vertex_count,
+                   u32 base_instance, u32 num_instances);
+
+    void DrawIndex(PrimitiveTopology topology, u32 index_first, u32 index_count, u32 base_index,
+                   u32 base_instance, u32 num_instances);
+
+    const State& GetDrawState() const {
+        return draw_state;
+    }
+
+private:
+    void SetInlineIndexBuffer(u32 index);
+
+    void DrawBegin();
+
+    void DrawEnd(u32 instance_count = 1, bool force_draw = false);
+
+    void DrawIndexSmall(u32 argument);
+
+    void ProcessTopologyOverride();
+
+    void ProcessDraw(bool draw_indexed, u32 instance_count);
+
+    Maxwell3D* maxwell3d{};
+    State draw_state{};
+    bool use_topology_override{};
+};
+} // namespace Tegra::Engines

src/video_core/engines/maxwell_3d.cpp

@@ -7,6 +7,7 @@
 #include "core/core.h"
 #include "core/core_timing.h"
 #include "video_core/dirty_flags.h"
+#include "video_core/engines/draw_manager.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
@@ -21,8 +22,10 @@ using VideoCore::QueryType;
 constexpr u32 MacroRegistersStart = 0xE00;
 
 Maxwell3D::Maxwell3D(Core::System& system_, MemoryManager& memory_manager_)
-    : system{system_}, memory_manager{memory_manager_}, macro_engine{GetMacroEngine(*this)},
+    : draw_manager{std::make_unique<DrawManager>(this)}, system{system_},
-      upload_state{memory_manager, regs.upload} {
+      memory_manager{memory_manager_}, macro_engine{GetMacroEngine(*this)}, upload_state{
+                                                                                memory_manager,
+                                                                                regs.upload} {
     dirty.flags.flip();
     InitializeRegisterDefaults();
 }
@@ -116,16 +119,6 @@ void Maxwell3D::InitializeRegisterDefaults() {
     regs.polygon_mode_front = Maxwell3D::Regs::PolygonMode::Fill;
 
     shadow_state = regs;
-
-    draw_command[MAXWELL3D_REG_INDEX(draw.end)] = true;
-    draw_command[MAXWELL3D_REG_INDEX(draw.begin)] = true;
-    draw_command[MAXWELL3D_REG_INDEX(vertex_buffer.first)] = true;
-    draw_command[MAXWELL3D_REG_INDEX(vertex_buffer.count)] = true;
-    draw_command[MAXWELL3D_REG_INDEX(index_buffer.first)] = true;
-    draw_command[MAXWELL3D_REG_INDEX(index_buffer.count)] = true;
-    draw_command[MAXWELL3D_REG_INDEX(draw_inline_index)] = true;
-    draw_command[MAXWELL3D_REG_INDEX(inline_index_2x16.even)] = true;
-    draw_command[MAXWELL3D_REG_INDEX(inline_index_4x8.index0)] = true;
 }
 
 void Maxwell3D::ProcessMacro(u32 method, const u32* base_start, u32 amount, bool is_last_call) {
@@ -213,29 +206,6 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
         return ProcessCBBind(3);
     case MAXWELL3D_REG_INDEX(bind_groups[4].raw_config):
         return ProcessCBBind(4);
-    case MAXWELL3D_REG_INDEX(index_buffer32_first):
-        regs.index_buffer.count = regs.index_buffer32_first.count;
-        regs.index_buffer.first = regs.index_buffer32_first.first;
-        dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
-        draw_indexed = true;
-        return ProcessDraw();
-    case MAXWELL3D_REG_INDEX(index_buffer16_first):
-        regs.index_buffer.count = regs.index_buffer16_first.count;
-        regs.index_buffer.first = regs.index_buffer16_first.first;
-        dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
-        draw_indexed = true;
-        return ProcessDraw();
-    case MAXWELL3D_REG_INDEX(index_buffer8_first):
-        regs.index_buffer.count = regs.index_buffer8_first.count;
-        regs.index_buffer.first = regs.index_buffer8_first.first;
-        dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
-        draw_indexed = true;
-        return ProcessDraw();
-    case MAXWELL3D_REG_INDEX(topology_override):
-        use_topology_override = true;
-        return;
-    case MAXWELL3D_REG_INDEX(clear_surface):
-        return ProcessClearBuffers(1);
     case MAXWELL3D_REG_INDEX(report_semaphore.query):
         return ProcessQueryGet();
     case MAXWELL3D_REG_INDEX(render_enable.mode):
@@ -254,6 +224,9 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
         return rasterizer->FragmentBarrier();
     case MAXWELL3D_REG_INDEX(tiled_cache_barrier):
         return rasterizer->TiledCacheBarrier();
+    default:
+        draw_manager->ProcessMethodCall(method, argument);
+        break;
     }
 }
 
@@ -268,7 +241,7 @@ void Maxwell3D::CallMacroMethod(u32 method, const std::vector<u32>& parameters)
     // Execute the current macro.
     macro_engine->Execute(macro_positions[entry], parameters);
 
-    ProcessDeferredDraw();
+    draw_manager->DrawDeferred();
 }
 
 void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
@@ -291,63 +264,8 @@ void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
     const u32 argument = ProcessShadowRam(method, method_argument);
     ProcessDirtyRegisters(method, argument);
 
-    if (draw_command[method]) {
-        regs.reg_array[method] = method_argument;
-        deferred_draw_method.push_back(method);
-        auto update_inline_index = [&](const u32 index) {
-            inline_index_draw_indexes.push_back(static_cast<u8>(index & 0x000000ff));
-            inline_index_draw_indexes.push_back(static_cast<u8>((index & 0x0000ff00) >> 8));
-            inline_index_draw_indexes.push_back(static_cast<u8>((index & 0x00ff0000) >> 16));
-            inline_index_draw_indexes.push_back(static_cast<u8>((index & 0xff000000) >> 24));
-            draw_mode = DrawMode::InlineIndex;
-        };
-        switch (method) {
-        case MAXWELL3D_REG_INDEX(draw.begin): {
-            draw_mode =
-                (regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Subsequent) ||
-                        (regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Unchanged)
-                    ? DrawMode::Instance
-                    : DrawMode::General;
-            break;
-        }
-        case MAXWELL3D_REG_INDEX(draw.end):
-            switch (draw_mode) {
-            case DrawMode::General:
-                ProcessDraw();
-                break;
-            case DrawMode::InlineIndex:
-                regs.index_buffer.count = static_cast<u32>(inline_index_draw_indexes.size() / 4);
-                regs.index_buffer.format = Regs::IndexFormat::UnsignedInt;
-                draw_indexed = true;
-                ProcessDraw();
-                inline_index_draw_indexes.clear();
-                break;
-            case DrawMode::Instance:
-                break;
-            }
-            break;
-        case MAXWELL3D_REG_INDEX(index_buffer.count):
-            draw_indexed = true;
-            break;
-        case MAXWELL3D_REG_INDEX(draw_inline_index):
-            update_inline_index(method_argument);
-            break;
-        case MAXWELL3D_REG_INDEX(inline_index_2x16.even):
-            update_inline_index(regs.inline_index_2x16.even);
-            update_inline_index(regs.inline_index_2x16.odd);
-            break;
-        case MAXWELL3D_REG_INDEX(inline_index_4x8.index0):
-            update_inline_index(regs.inline_index_4x8.index0);
-            update_inline_index(regs.inline_index_4x8.index1);
-            update_inline_index(regs.inline_index_4x8.index2);
-            update_inline_index(regs.inline_index_4x8.index3);
-            break;
-        }
-    } else {
-        ProcessDeferredDraw();
     ProcessMethodCall(method, argument, method_argument, is_last_call);
 }
-}
 
 void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
                                 u32 methods_pending) {
@@ -387,35 +305,6 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
     }
 }
 
-void Maxwell3D::ProcessTopologyOverride() {
-    using PrimitiveTopology = Maxwell3D::Regs::PrimitiveTopology;
-    using PrimitiveTopologyOverride = Maxwell3D::Regs::PrimitiveTopologyOverride;
-
-    PrimitiveTopology topology{};
-
-    switch (regs.topology_override) {
-    case PrimitiveTopologyOverride::None:
-        topology = regs.draw.topology;
-        break;
-    case PrimitiveTopologyOverride::Points:
-        topology = PrimitiveTopology::Points;
-        break;
-    case PrimitiveTopologyOverride::Lines:
-        topology = PrimitiveTopology::Lines;
-        break;
-    case PrimitiveTopologyOverride::LineStrip:
-        topology = PrimitiveTopology::LineStrip;
-        break;
-    default:
-        topology = static_cast<PrimitiveTopology>(regs.topology_override);
-        break;
-    }
-
-    if (use_topology_override) {
-        regs.draw.topology.Assign(topology);
-    }
-}
-
 void Maxwell3D::ProcessMacroUpload(u32 data) {
     macro_engine->AddCode(regs.load_mme.instruction_ptr++, data);
 }
@@ -625,44 +514,4 @@ u32 Maxwell3D::GetRegisterValue(u32 method) const {
     return regs.reg_array[method];
 }
 
-void Maxwell3D::ProcessClearBuffers(u32 layer_count) {
-    rasterizer->Clear(layer_count);
-}
-
-void Maxwell3D::ProcessDraw(u32 instance_count) {
-    LOG_TRACE(HW_GPU, "called, topology={}, count={}", regs.draw.topology.Value(),
-              draw_indexed ? regs.index_buffer.count : regs.vertex_buffer.count);
-
-    ProcessTopologyOverride();
-
-    if (ShouldExecute()) {
-        rasterizer->Draw(draw_indexed, instance_count);
-    }
-
-    draw_indexed = false;
-    deferred_draw_method.clear();
-}
-
-void Maxwell3D::ProcessDeferredDraw() {
-    if (draw_mode != DrawMode::Instance || deferred_draw_method.empty()) {
-        return;
-    }
-
-    const auto method_count = deferred_draw_method.size();
-    u32 instance_count = 1;
-    u32 vertex_buffer_count = 0;
-    u32 index_buffer_count = 0;
-    for (size_t index = 0; index < method_count; ++index) {
-        const u32 method = deferred_draw_method[index];
-        if (method == MAXWELL3D_REG_INDEX(vertex_buffer.count)) {
-            instance_count = ++vertex_buffer_count;
-        } else if (method == MAXWELL3D_REG_INDEX(index_buffer.count)) {
-            instance_count = ++index_buffer_count;
-        }
-    }
-    ASSERT_MSG(!(vertex_buffer_count && index_buffer_count), "Instance both indexed and direct?");
-
-    ProcessDraw(instance_count);
-}
-
 } // namespace Tegra::Engines

src/video_core/engines/maxwell_3d.h

@@ -37,6 +37,8 @@ class RasterizerInterface;
 
 namespace Tegra::Engines {
 
+class DrawManager;
+
 /**
  * This Engine is known as GF100_3D. Documentation can be found in:
  * https://github.com/NVIDIA/open-gpu-doc/blob/master/classes/3d/clb197.h
@@ -2223,6 +2225,7 @@ public:
 
         struct IndexBufferSmall {
             union {
+                u32 raw;
                 BitField<0, 16, u32> first;
                 BitField<16, 12, u32> count;
                 BitField<28, 4, PrimitiveTopology> topology;
@@ -3061,10 +3064,8 @@ public:
         Tables tables{};
     } dirty;
 
-    std::vector<u8> inline_index_draw_indexes;
+    std::unique_ptr<DrawManager> draw_manager;
-
+    friend class DrawManager;
-    /// Handles a write to the CLEAR_BUFFERS register.
-    void ProcessClearBuffers(u32 layer_count);
 
 private:
     void InitializeRegisterDefaults();
@@ -3122,15 +3123,6 @@ private:
     /// Handles a write to the CB_BIND register.
     void ProcessCBBind(size_t stage_index);
 
-    /// Handles use of topology overrides (e.g., to avoid using a topology assigned from a macro)
-    void ProcessTopologyOverride();
-
-    /// Handles deferred draw(e.g., instance draw).
-    void ProcessDeferredDraw();
-
-    /// Handles a draw.
-    void ProcessDraw(u32 instance_count = 1);
-
     /// Returns a query's value or an empty object if the value will be deferred through a cache.
     std::optional<u64> GetQueryResult();
 
@@ -3153,13 +3145,6 @@ private:
     Upload::State upload_state;
 
     bool execute_on{true};
-    bool use_topology_override{false};
-
-    std::array<bool, Regs::NUM_REGS> draw_command{};
-    std::vector<u32> deferred_draw_method;
-    enum class DrawMode : u32 { General = 0, Instance, InlineIndex };
-    DrawMode draw_mode{DrawMode::General};
-    bool draw_indexed{};
 };
 
 #define ASSERT_REG_POSITION(field_name, position)                                                  \

src/video_core/macro/macro_hle.cpp

@@ -5,6 +5,7 @@
 #include <vector>
 #include "common/scope_exit.h"
 #include "video_core/dirty_flags.h"
+#include "video_core/engines/draw_manager.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/macro/macro.h"
 #include "video_core/macro/macro_hle.h"
@@ -18,57 +19,33 @@ using HLEFunction = void (*)(Engines::Maxwell3D& maxwell3d, const std::vector<u3
 // HLE'd functions
 void HLE_771BB18C62444DA0(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters) {
     const u32 instance_count = parameters[2] & maxwell3d.GetRegisterValue(0xD1B);
-
+    maxwell3d.draw_manager->DrawIndex(
-    maxwell3d.regs.draw.topology.Assign(
+        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0] & 0x3ffffff),
-        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0] & 0x3ffffff));
+        parameters[4], parameters[1], parameters[3], parameters[5], instance_count);
-    maxwell3d.regs.global_base_instance_index = parameters[5];
-    maxwell3d.regs.global_base_vertex_index = parameters[3];
-    maxwell3d.regs.index_buffer.count = parameters[1];
-    maxwell3d.regs.index_buffer.first = parameters[4];
-
-    if (maxwell3d.ShouldExecute()) {
-        maxwell3d.Rasterizer().Draw(true, instance_count);
-    }
-    maxwell3d.regs.index_buffer.count = 0;
 }
 
 void HLE_0D61FC9FAAC9FCAD(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters) {
     const u32 instance_count = (maxwell3d.GetRegisterValue(0xD1B) & parameters[2]);
-
+    maxwell3d.draw_manager->DrawArray(
-    maxwell3d.regs.vertex_buffer.first = parameters[3];
+        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]),
-    maxwell3d.regs.vertex_buffer.count = parameters[1];
+        parameters[3], parameters[1], parameters[4], instance_count);
-    maxwell3d.regs.global_base_instance_index = parameters[4];
-    maxwell3d.regs.draw.topology.Assign(
-        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]));
-
-    if (maxwell3d.ShouldExecute()) {
-        maxwell3d.Rasterizer().Draw(false, instance_count);
-    }
-    maxwell3d.regs.vertex_buffer.count = 0;
 }
 
 void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters) {
     const u32 instance_count = (maxwell3d.GetRegisterValue(0xD1B) & parameters[2]);
     const u32 element_base = parameters[4];
     const u32 base_instance = parameters[5];
-    maxwell3d.regs.index_buffer.first = parameters[3];
     maxwell3d.regs.vertex_id_base = element_base;
-    maxwell3d.regs.index_buffer.count = parameters[1];
     maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
-    maxwell3d.regs.global_base_vertex_index = element_base;
-    maxwell3d.regs.global_base_instance_index = base_instance;
     maxwell3d.CallMethod(0x8e3, 0x640, true);
     maxwell3d.CallMethod(0x8e4, element_base, true);
     maxwell3d.CallMethod(0x8e5, base_instance, true);
-    maxwell3d.regs.draw.topology.Assign(
+
-        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]));
+    maxwell3d.draw_manager->DrawIndex(
-    if (maxwell3d.ShouldExecute()) {
+        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]),
-        maxwell3d.Rasterizer().Draw(true, instance_count);
+        parameters[3], parameters[1], element_base, base_instance, instance_count);
-    }
+
     maxwell3d.regs.vertex_id_base = 0x0;
-    maxwell3d.regs.index_buffer.count = 0;
-    maxwell3d.regs.global_base_vertex_index = 0x0;
-    maxwell3d.regs.global_base_instance_index = 0x0;
     maxwell3d.CallMethod(0x8e3, 0x640, true);
     maxwell3d.CallMethod(0x8e4, 0x0, true);
     maxwell3d.CallMethod(0x8e5, 0x0, true);
@@ -79,9 +56,6 @@ void HLE_3F5E74B9C9A50164(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     SCOPE_EXIT({
         // Clean everything.
         maxwell3d.regs.vertex_id_base = 0x0;
-        maxwell3d.regs.index_buffer.count = 0;
-        maxwell3d.regs.global_base_vertex_index = 0x0;
-        maxwell3d.regs.global_base_instance_index = 0x0;
         maxwell3d.CallMethod(0x8e3, 0x640, true);
         maxwell3d.CallMethod(0x8e4, 0x0, true);
         maxwell3d.CallMethod(0x8e5, 0x0, true);
@@ -93,9 +67,6 @@ void HLE_3F5E74B9C9A50164(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
         // Nothing to do.
         return;
     }
-    const auto topology =
-        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[2]);
-    maxwell3d.regs.draw.topology.Assign(topology);
     const u32 padding = parameters[3];
     const std::size_t max_draws = parameters[4];
 
@@ -106,23 +77,17 @@ void HLE_3F5E74B9C9A50164(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
 
     for (std::size_t index = first_draw; index < last_draw; index++) {
         const std::size_t base = index * indirect_words + 5;
-        const u32 num_vertices = parameters[base];
-        const u32 instance_count = parameters[base + 1];
-        const u32 first_index = parameters[base + 2];
         const u32 base_vertex = parameters[base + 3];
         const u32 base_instance = parameters[base + 4];
-        maxwell3d.regs.index_buffer.first = first_index;
         maxwell3d.regs.vertex_id_base = base_vertex;
-        maxwell3d.regs.index_buffer.count = num_vertices;
-        maxwell3d.regs.global_base_vertex_index = base_vertex;
-        maxwell3d.regs.global_base_instance_index = base_instance;
         maxwell3d.CallMethod(0x8e3, 0x640, true);
         maxwell3d.CallMethod(0x8e4, base_vertex, true);
         maxwell3d.CallMethod(0x8e5, base_instance, true);
         maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
-        if (maxwell3d.ShouldExecute()) {
+        maxwell3d.draw_manager->DrawIndex(
-            maxwell3d.Rasterizer().Draw(true, instance_count);
+            static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[2]),
-        }
+            parameters[base + 2], parameters[base], base_vertex, base_instance,
+            parameters[base + 1]);
     }
 }
 
@@ -136,7 +101,7 @@ void HLE_EAD26C3E2109B06B(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     ASSERT(clear_params.layer == 0);
 
     maxwell3d.regs.clear_surface.raw = clear_params.raw;
-    maxwell3d.ProcessClearBuffers(num_layers);
+    maxwell3d.draw_manager->Clear(num_layers);
 }
 
 constexpr std::array<std::pair<u64, HLEFunction>, 5> hle_funcs{{

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -224,16 +224,18 @@ void RasterizerOpenGL::Draw(bool is_indexed, u32 instance_count) {
 
     SyncState();
 
-    const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d->regs.draw.topology);
+    const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
+
+    const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(draw_state.topology);
     BeginTransformFeedback(pipeline, primitive_mode);
 
-    const GLuint base_instance = static_cast<GLuint>(maxwell3d->regs.global_base_instance_index);
+    const GLuint base_instance = static_cast<GLuint>(draw_state.base_instance);
     const GLsizei num_instances = static_cast<GLsizei>(instance_count);
     if (is_indexed) {
-        const GLint base_vertex = static_cast<GLint>(maxwell3d->regs.global_base_vertex_index);
+        const GLint base_vertex = static_cast<GLint>(draw_state.base_index);
-        const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d->regs.index_buffer.count);
+        const GLsizei num_vertices = static_cast<GLsizei>(draw_state.index_buffer.count);
         const GLvoid* const offset = buffer_cache_runtime.IndexOffset();
-        const GLenum format = MaxwellToGL::IndexFormat(maxwell3d->regs.index_buffer.format);
+        const GLenum format = MaxwellToGL::IndexFormat(draw_state.index_buffer.format);
         if (num_instances == 1 && base_instance == 0 && base_vertex == 0) {
             glDrawElements(primitive_mode, num_vertices, format, offset);
         } else if (num_instances == 1 && base_instance == 0) {
@@ -252,8 +254,8 @@ void RasterizerOpenGL::Draw(bool is_indexed, u32 instance_count) {
                                                           base_instance);
         }
     } else {
-        const GLint base_vertex = static_cast<GLint>(maxwell3d->regs.vertex_buffer.first);
+        const GLint base_vertex = static_cast<GLint>(draw_state.vertex_buffer.first);
-        const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d->regs.vertex_buffer.count);
+        const GLsizei num_vertices = static_cast<GLsizei>(draw_state.vertex_buffer.count);
         if (num_instances == 1 && base_instance == 0) {
             glDrawArrays(primitive_mode, base_vertex, num_vertices);
         } else if (base_instance == 0) {

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -22,6 +22,7 @@
 #include "shader_recompiler/frontend/maxwell/control_flow.h"
 #include "shader_recompiler/frontend/maxwell/translate_program.h"
 #include "shader_recompiler/profile.h"
+#include "video_core/engines/draw_manager.h"
 #include "video_core/engines/kepler_compute.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/memory_manager.h"
@@ -327,7 +328,7 @@ GraphicsPipeline* ShaderCache::CurrentGraphicsPipeline() {
     const auto& regs{maxwell3d->regs};
     graphics_key.raw = 0;
     graphics_key.early_z.Assign(regs.mandated_early_z != 0 ? 1 : 0);
-    graphics_key.gs_input_topology.Assign(regs.draw.topology.Value());
+    graphics_key.gs_input_topology.Assign(maxwell3d->draw_manager->GetDrawState().topology);
     graphics_key.tessellation_primitive.Assign(regs.tessellation.params.domain_type.Value());
     graphics_key.tessellation_spacing.Assign(regs.tessellation.params.spacing.Value());
     graphics_key.tessellation_clockwise.Assign(
@@ -371,7 +372,8 @@ GraphicsPipeline* ShaderCache::BuiltPipeline(GraphicsPipeline* pipeline) const n
     // If games are using a small index count, we can assume these are full screen quads.
     // Usually these shaders are only used once for building textures so we can assume they
     // can't be built async
-    if (maxwell3d->regs.index_buffer.count <= 6 || maxwell3d->regs.vertex_buffer.count <= 6) {
+    const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
+    if (draw_state.index_buffer.count <= 6 || draw_state.vertex_buffer.count <= 6) {
         return pipeline;
     }
     return nullptr;

src/video_core/renderer_vulkan/fixed_pipeline_state.cpp

@@ -8,6 +8,7 @@
 #include "common/cityhash.h"
 #include "common/common_types.h"
 #include "common/polyfill_ranges.h"
+#include "video_core/engines/draw_manager.h"
 #include "video_core/renderer_vulkan/fixed_pipeline_state.h"
 #include "video_core/renderer_vulkan/vk_state_tracker.h"
 
@@ -50,12 +51,13 @@ void RefreshXfbState(VideoCommon::TransformFeedbackState& state, const Maxwell&
 void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d,
                                  bool has_extended_dynamic_state, bool has_dynamic_vertex_input) {
     const Maxwell& regs = maxwell3d.regs;
+    const auto topology_ = maxwell3d.draw_manager->GetDrawState().topology;
     const std::array enabled_lut{
         regs.polygon_offset_point_enable,
         regs.polygon_offset_line_enable,
         regs.polygon_offset_fill_enable,
     };
-    const u32 topology_index = static_cast<u32>(regs.draw.topology.Value());
+    const u32 topology_index = static_cast<u32>(topology_);
 
     raw1 = 0;
     extended_dynamic_state.Assign(has_extended_dynamic_state ? 1 : 0);
@@ -78,7 +80,7 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d,
                                   Maxwell::Tessellation::OutputPrimitives::Triangles_CW);
     logic_op_enable.Assign(regs.logic_op.enable != 0 ? 1 : 0);
     logic_op.Assign(PackLogicOp(regs.logic_op.op));
-    topology.Assign(regs.draw.topology);
+    topology.Assign(topology_);
     msaa_mode.Assign(regs.anti_alias_samples_mode);
 
     raw2 = 0;

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -507,7 +507,8 @@ GraphicsPipeline* PipelineCache::BuiltPipeline(GraphicsPipeline* pipeline) const
     // If games are using a small index count, we can assume these are full screen quads.
     // Usually these shaders are only used once for building textures so we can assume they
     // can't be built async
-    if (maxwell3d->regs.index_buffer.count <= 6 || maxwell3d->regs.vertex_buffer.count <= 6) {
+    const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
+    if (draw_state.index_buffer.count <= 6 || draw_state.vertex_buffer.count <= 6) {
         return pipeline;
     }
     return nullptr;

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -12,6 +12,7 @@
 #include "common/scope_exit.h"
 #include "common/settings.h"
 #include "video_core/control/channel_state.h"
+#include "video_core/engines/draw_manager.h"
 #include "video_core/engines/kepler_compute.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/blit_image.h"
@@ -36,6 +37,7 @@
 namespace Vulkan {
 
 using Maxwell = Tegra::Engines::Maxwell3D::Regs;
+using MaxwellDrawState = Tegra::Engines::DrawManager::State;
 using VideoCommon::ImageViewId;
 using VideoCommon::ImageViewType;
 
@@ -127,16 +129,16 @@ VkRect2D GetScissorState(const Maxwell& regs, size_t index, u32 up_scale = 1, u3
     return scissor;
 }
 
-DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_indexed) {
+DrawParams MakeDrawParams(const MaxwellDrawState& draw_state, u32 num_instances, bool is_indexed) {
     DrawParams params{
-        .base_instance = regs.global_base_instance_index,
+        .base_instance = draw_state.base_instance,
         .num_instances = num_instances,
-        .base_vertex = is_indexed ? regs.global_base_vertex_index : regs.vertex_buffer.first,
+        .base_vertex = is_indexed ? draw_state.base_index : draw_state.vertex_buffer.first,
-        .num_vertices = is_indexed ? regs.index_buffer.count : regs.vertex_buffer.count,
+        .num_vertices = is_indexed ? draw_state.index_buffer.count : draw_state.vertex_buffer.count,
-        .first_index = is_indexed ? regs.index_buffer.first : 0,
+        .first_index = is_indexed ? draw_state.index_buffer.first : 0,
         .is_indexed = is_indexed,
     };
-    if (regs.draw.topology == Maxwell::PrimitiveTopology::Quads) {
+    if (draw_state.topology == Maxwell::PrimitiveTopology::Quads) {
         // 6 triangle vertices per quad, base vertex is part of the index
         // See BindQuadArrayIndexBuffer for more details
         params.num_vertices = (params.num_vertices / 4) * 6;
@@ -195,9 +197,9 @@ void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) {
 
     UpdateDynamicStates();
 
-    const auto& regs{maxwell3d->regs};
+    const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
     const u32 num_instances{instance_count};
-    const DrawParams draw_params{MakeDrawParams(regs, num_instances, is_indexed)};
+    const DrawParams draw_params{MakeDrawParams(draw_state, num_instances, is_indexed)};
     scheduler.Record([draw_params](vk::CommandBuffer cmdbuf) {
         if (draw_params.is_indexed) {
             cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances,