Execute Windows DLSS K and M at recorded D3D12 command-list boundaries

CMakeLists.txt

@@ -1,5 +1,5 @@
 cmake_minimum_required(VERSION 3.24)
-project(d4r_windows_diagnostics LANGUAGES CXX)
+project(d4r_windows_diagnostics LANGUAGES C CXX)
 if(NOT WIN32 OR NOT CMAKE_SIZEOF_VOID_P EQUAL 8)
   message(FATAL_ERROR "These diagnostics require native Windows x64")
 endif()
@@ -68,7 +68,7 @@ foreach(family k m)
   endforeach()
 endforeach()
 add_custom_target(d4r_native_gfx1201_compile ALL DEPENDS ${native_objects})
-foreach(probe hip_gfx1201_probe gfx12_wmma_probe k_module_probe native_replay_probe cuda_driver_probe cuda_mma_probe cuda_image_probe d3d12_hip_interop_probe hip_stream_lifecycle_probe ngx_cuda_init_probe d3d12_dlss_probe)
+foreach(probe hip_gfx1201_probe gfx12_wmma_probe k_module_probe native_replay_probe cuda_driver_probe cuda_mma_probe cuda_image_probe d3d12_command_probe d3d12_hip_interop_probe hip_stream_lifecycle_probe ngx_cuda_init_probe d3d12_dlss_probe)
   add_executable(d4r_${probe} tools/windows/${probe}.cpp)
   target_compile_definitions(d4r_${probe} PRIVATE WIN32_LEAN_AND_MEAN NOMINMAX __HIP_PLATFORM_AMD__)
   target_include_directories(d4r_${probe} PRIVATE "${D4R_HIP_ROOT}/include")
@@ -84,11 +84,17 @@ foreach(probe hip_gfx1201_probe gfx12_wmma_probe k_module_probe native_replay_pr
   endif()
   install(TARGETS d4r_${probe} RUNTIME DESTINATION bin)
 endforeach()
-add_library(d4r_nvngx_windows SHARED tools/windows/d4r_nvngx_windows.cpp)
+set(D4R_MINHOOK_ROOT "${CMAKE_CURRENT_SOURCE_DIR}/external/MinHook" CACHE PATH "MinHook 1.3.4 source checkout")
+if(NOT EXISTS "${D4R_MINHOOK_ROOT}/include/MinHook.h")
+  message(FATAL_ERROR "Clone MinHook v1.3.4 into external/MinHook or set D4R_MINHOOK_ROOT")
+endif()
+add_subdirectory("${D4R_MINHOOK_ROOT}" "${CMAKE_CURRENT_BINARY_DIR}/minhook" EXCLUDE_FROM_ALL)
+add_library(d4r_nvngx_windows SHARED tools/windows/d4r_nvngx_windows.cpp tools/windows/d3d12_command_hooks.cpp)
 set_target_properties(d4r_nvngx_windows PROPERTIES PREFIX "" OUTPUT_NAME "d4r_nvngx")
 target_compile_definitions(d4r_nvngx_windows PRIVATE WIN32_LEAN_AND_MEAN NOMINMAX __HIP_PLATFORM_AMD__)
 target_include_directories(d4r_nvngx_windows PRIVATE "${D4R_HIP_ROOT}/include")
-target_link_libraries(d4r_nvngx_windows PRIVATE version dbghelp d3d12 dxguid)
+target_link_libraries(d4r_nvngx_windows PRIVATE version dbghelp d3d12 dxguid minhook)
+install(FILES "${D4R_MINHOOK_ROOT}/LICENSE.txt" DESTINATION licenses RENAME MinHook.txt)
 if(MINGW)
   target_link_options(d4r_nvngx_windows PRIVATE -static)
 endif()
@@ -132,6 +138,8 @@ endif()
 install(TARGETS d4r_nvapi_compat RUNTIME DESTINATION nvapi-compat)
 target_link_libraries(d4r_d3d12_hip_interop_probe PRIVATE d3d12 dxgi dxguid)
 target_link_libraries(d4r_d3d12_dlss_probe PRIVATE d3d12 dxgi dxguid)
+target_link_libraries(d4r_d3d12_command_probe PRIVATE d3d12 dxgi dxguid)
+add_dependencies(d4r_d3d12_command_probe d4r_nvngx_windows)
 add_dependencies(d4r_hip_gfx1201_probe d4r_probe_code_object)
 add_dependencies(d4r_gfx12_wmma_probe d4r_wmma_code_object)
 add_dependencies(d4r_k_module_probe d4r_native_gfx1201_compile)
@@ -155,6 +163,9 @@ install(FILES kernels/tools/pwin_model.py kernels/tools/pwin_check.py kernels/to
 install(FILES kernels/tools/model_enc3.py kernels/tools/swin_model.py kernels/tools/swin_check.py DESTINATION reference)
 install(FILES docs/windows-rdna4-port.md DESTINATION docs)
 include(CTest)
+add_test(NAME d3d12_command_backend COMMAND d4r_d3d12_command_probe
+  --hip-root "${D4R_HIP_ROOT}" --module "$<TARGET_FILE:d4r_nvngx_windows>" --iterations 32)
+set_tests_properties(d3d12_command_backend PROPERTIES TIMEOUT 120 LABELS "hardware;d3d12;command_lists;lifetime")
 add_test(NAME hip_gfx1201 COMMAND d4r_hip_gfx1201_probe
   --hip-root "${D4R_HIP_ROOT}" --module "${probe_object}")
 set_tests_properties(hip_gfx1201 PROPERTIES TIMEOUT 90 LABELS "hardware;gfx1201")

docs/windows-rdna4-port.md

@@ -13,7 +13,10 @@ frames on Windows. M now passes all five synthetic and real-weight layers agains
 NumPy and independent PTX, and four native CUDA frames. The standalone D3D12
 harness now produces four K and four M frames using imported VRAM and shared
 fences, with bit-exact RGB agreement against the corresponding CUDA harnesses.
-Game command-list/queue integration remains in progress. A full temporal M scan
+Game integration remains in progress. The ordinary NGX D3D12 Evaluate API now
+records a boundary in an open command list; a native queue interception backend
+executes prefix, CUDA evaluation, then suffix in the same submitted frame.
+A full temporal M scan
 found rare attention P/V accumulation mismatches that the earlier twelve-block
 sampling missed. `SWIN_EXACT_PV` fixes the native baseline; ZLUDA patch `0013`
 adds the corresponding opt-in f16 reference accumulation. All spatial blocks in
@@ -31,7 +34,7 @@ Do not integrate NGX until the integer PTX workload is stable on the real GPU.
 | M5: independently validated gfx12 WMMA backend | PASS: raw + legacy adapter + upstream layout, max abs/relative error 0 |
 | M6: K layers, full transformer and image validation | PASS: 11 real-weight layers, 44 native launches, four D3D12 frames bit-exact against native CUDA, 81.85–92.91 dB versus translated K |
 | M7: M FP16-equivalent baseline and full transformer | PASS: all blocks of 40 native launches exactly match NumPy; targeted independent PTX regression matches; four D3D12 frames bit-exact against corrected translated M |
-| M8: standalone Windows D3D12 NGX harness | PASS at explicit submitted queue boundaries: K/M, four frames each, no shim CPU image copies, bit-exact RGB against CUDA harness |
+| M8: standalone Windows D3D12 NGX harness | PASS for ordinary recorded EvaluateFeature and explicit boundaries: K/M, four frames each, no shim CPU image copies; both bit-exact against explicit boundary baselines |
 | M9: OptiScaler integration, profiling, installation | Pending |
 
 ## M temporal precision regression (2026-09-30)
@@ -93,6 +96,57 @@ The game selected for the next integration test is Silent Hill 2 at
 `D:\Games\SILENT HILL 2\SHProto\Binaries\Win64\SHProto-Win64-Shipping.exe`.
 No game files have been modified at this milestone.
 
+## Open D3D12 command lists (2026-09-30)
+
+The backend intercepts documented D3D12 COM methods using pinned open-source
+[MinHook 1.3.4](https://github.com/TsudaKageyu/minhook/releases/tag/v1.3.4), commit
+`c3fcafdc10146beb5919319d0683e44e3c30d537`. No NVIDIA instructions, proprietary
+DLL contents or private driver interfaces are patched. The typed forwarding
+code is generated from public SDK CommandList0..7 declarations. The additional
+CommandList8..10 declarations were audited in Microsoft's current public
+DirectX-Headers, `adbd6f3ba40795c46a8d0f33af00bcb57ff0f0a4`; coverage of those
+newer methods is the next implementation step.
+
+At NGX Evaluate, the caller's resource references and typed parameter values are
+copied into an owned snapshot. The recorded prefix is closed. Subsequent calls
+on the same logical object are forwarded into a fresh native suffix list with
+its own allocator. State-setting calls are replayed with deep copies of host
+arrays and retained COM references. Open PIX events are closed in the prefix
+and reopened in the suffix. At ExecuteCommandLists, preceding lists in the
+caller's batch execute first, followed by the prefix, actual CUDA evaluation
+and output copy, then the suffix and following lists. All work completes in the
+same frame. Completion waits currently stall the host; they do not age frames.
+
+The callback module remains loaded while public D3D12 method interceptions are
+installed. Original command-list pointers in the routing map are weak; final
+Release removes metadata, suffixes, allocators and queued feature references.
+This avoids a reference cycle that would keep every command list alive.
+Failure at queue submission is logged and prevents submitting the consumer
+suffix. No previous-frame fallback is substituted.
+
+`d3d12-command-foundation` passes 32 iterations with no NVIDIA binaries. Each
+batch contains A, B and C. A produces shader inputs, B dispatches before a
+boundary, the callback verifies that prefix and replaces inputs on the GPU,
+the suffix dispatches without rebinding root state, and C reads the result.
+All values match; a temporary root-constant array may change after recording;
+live recording metadata returns to zero after every iteration.
+
+`d3d12-m-command-list.zip` passes four ordinary NGX EvaluateFeature frames with
+forty native M launches. RGB is bit-exact to the earlier explicit boundary
+harness. The initial producer remains unsubmitted until the same queue batch,
+and the caller's Color/Reset parameters are deliberately changed after
+Evaluate returns, testing snapshot ownership. `d3d12-k-command-list.zip` also
+passes four finite frames and all forty-four native K launches; all four RGB
+frames are bit-exact against `d3d12-k-final`. Complete CTest: 12/12 PASS
+(`d3d12-command-final-ctest.log`).
+
+Enable this backend with `D4R_D3D12_COMMAND_BACKEND=1`; the diagnostic runner
+sets it through `-CommandListBackend`. Before game testing, remaining gates are
+newer command methods, indirect-state reconstruction, enhanced barrier layouts,
+actual game texture formats and interaction with OptiScaler's own state hooks.
+Closed lists, active render/query scopes, GPU predication and unknown indirect
+state are currently explicitly rejected instead of guessing their semantics.
+
 ## Source baseline (checked 2026-09-29)
 
 * d4r: `ed1ab60`, main, 14 commits; https://github.com/countervolts/d4r.

scripts/windows/build-windows-rdna4.ps1

@@ -35,6 +35,15 @@ if (!(Test-Path $cmake) -or !(Test-Path $ninja)) {
     throw 'CMake >=3.24 and Ninja required. See docs/windows-rdna4-port.md for local tool setup.'
 }
 New-Item -ItemType Directory -Force $BuildDirectory | Out-Null
+$minhook = Join-Path $repo 'external/MinHook'
+if (!(Test-Path (Join-Path $minhook 'include/MinHook.h'))) {
+    & git clone --depth 1 --branch v1.3.4 https://github.com/TsudaKageyu/minhook.git $minhook
+    if ($LASTEXITCODE) { throw 'MinHook 1.3.4 source download failed.' }
+}
+$minhookCommit = (& git -C $minhook rev-parse HEAD).Trim()
+if ($minhookCommit -ne 'c3fcafdc10146beb5919319d0683e44e3c30d537') {
+    throw "Expected audited MinHook 1.3.4 source, found $minhookCommit"
+}
 $originalPath = $env:PATH
 try {
     $configure = @('--fresh', '-S', $repo, '-B', $BuildDirectory, '-G', 'Ninja', '-DCMAKE_BUILD_TYPE=RelWithDebInfo',
@@ -44,6 +53,7 @@ try {
         if (!(Test-Path $compiler)) { throw "Compiler missing: $compiler" }
         $env:PATH = "$(Join-Path $ToolchainRoot 'bin');$originalPath"
         $configure += "-DCMAKE_CXX_COMPILER=$compiler"
+        $configure += "-DCMAKE_C_COMPILER=$(Join-Path $ToolchainRoot 'bin/x86_64-w64-mingw32-clang.exe')"
     }
     if ($ZludaRoot) { $configure += "-DD4R_ZLUDA_ROOT=$ZludaRoot" }
     $ErrorActionPreference = 'Continue'

scripts/windows/test-windows-rdna4.ps1

@@ -12,6 +12,7 @@ param(
     [switch]$NgxOnly,
     [switch]$Trace,
     [switch]$RequireNativeNetwork,
+    [switch]$CommandListBackend,
     [int]$Iterations = 32,
     [int]$TimeoutSeconds = 180
 )
@@ -210,6 +211,9 @@ try {
         if ($cudaOk) {
             $mapOk = $interopOk = $true
             if (!$NgxOnly) {
+            $commandOk = Invoke-Probe 'd3d12-command-backend' (Join-Path $bin 'd4r_d3d12_command_probe.exe') @(
+                '--hip-root', $HipRoot, '--module', (Join-Path $bin 'd4r_nvngx.dll'), '--iterations', "$Iterations")
+            if (!$commandOk) { throw 'D3D12 command-list order/state/lifetime validation failed.' }
             $mapOk = Invoke-Probe 'interop-map-lifetime' (Join-Path $bin 'd4r_d3d12_hip_interop_probe.exe') @(
                 '--hip-root', $HipRoot, '--interop-mode', 'map', '--iterations', '64')
             $interopOk = Invoke-Probe 'interop-roundtrip' (Join-Path $bin 'd4r_d3d12_hip_interop_probe.exe') @(
@@ -250,6 +254,7 @@ try {
                     $localShim = Join-Path $ngxRuntimeDirectory 'd4r_nvngx.dll'
                     Copy-Item -LiteralPath (Join-Path $bin 'd4r_nvngx.dll') -Destination $localShim
                     $ngxArguments += @('--module', $localShim)
+                    if ($CommandListBackend) { $ngxArguments += @('--interop-mode', 'command-list') }
                 }
                 $ngxName = if ($NgxMode -eq 'evaluate') { "ngx-evaluate-preset-$Preset" } elseif ($NgxMode -eq 'd3d12') { "d3d12-evaluate-preset-$Preset" } else { 'ngx-init' }
                 $ngxOk = Invoke-Probe $ngxName $ngxExe $ngxArguments

tools/windows/d3d12_command_hooks.cpp

@@ -0,0 +1,320 @@
+#include "d3d12_command_hooks.h"
+#include <MinHook.h>
+#include <atomic>
+#include <memory>
+#include <mutex>
+#include <unordered_map>
+#include <vector>
+
+namespace d4r::win::commands {
+namespace {
+thread_local bool internal = false;
+std::mutex map_mutex, installation_mutex, error_mutex;
+std::string last_error;
+bool installed = false;
+struct Event { UINT metadata; std::vector<uint8_t> data; };
+using State = std::function<void(ID3D12GraphicsCommandList*)>;
+struct Segment {
+    // The application's original list is a weak pointer; holding it in its
+    // own routing map would prevent final Release and leak the entire feature.
+    ID3D12GraphicsCommandList* list = nullptr;
+    ComPtr<ID3D12GraphicsCommandList> owner;
+    ComPtr<ID3D12CommandAllocator> allocator;
+    Boundary boundary;
+};
+struct Recording {
+    std::recursive_mutex mutex;
+    ID3D12GraphicsCommandList* original;
+    ComPtr<ID3D12GraphicsCommandList> current;
+    ComPtr<ID3D12CommandAllocator> current_allocator;
+    std::vector<Segment> segments;
+    std::vector<State> state;
+    std::vector<Event> events;
+    std::unordered_map<ID3D12Resource*, D3D12_RESOURCE_STATES> resource_states;
+    unsigned active_queries = 0;
+    bool render_pass = false, predication = false, indirect_state_unknown = false, closed = false;
+    explicit Recording(ID3D12GraphicsCommandList* list) : original(list) {}
+};
+std::unordered_map<ID3D12GraphicsCommandList*, std::shared_ptr<Recording>> recordings;
+std::shared_ptr<Recording> get_recording(ID3D12GraphicsCommandList* self, bool create = true) {
+    if (internal) return {};
+    std::lock_guard<std::mutex> lock(map_mutex);
+    auto found = recordings.find(self);
+    if (found != recordings.end()) return found->second;
+    if (!create) return {};
+    auto value = std::make_shared<Recording>(self);
+    recordings.emplace(self, value); return value;
+}
+void forget(ID3D12GraphicsCommandList* self) {
+    std::shared_ptr<Recording> released;
+    { std::lock_guard<std::mutex> lock(map_mutex);
+      auto it = recordings.find(self);
+      if (it != recordings.end()) { released = std::move(it->second); recordings.erase(it); } }
+    // Release retained COM objects without holding the map mutex.
+}
+struct Access {
+    std::shared_ptr<Recording> recording;
+    std::unique_lock<std::recursive_mutex> lock;
+    ID3D12GraphicsCommandList* self;
+    explicit Access(ID3D12GraphicsCommandList* list) : recording(get_recording(list)), self(list) {
+        if (recording) lock = std::unique_lock<std::recursive_mutex>(recording->mutex);
+    }
+    ID3D12GraphicsCommandList* target() const {
+        return recording && recording->current ? recording->current.Get() : self;
+    }
+};
+template<class T> ComPtr<T> retain(T* pointer) { return ComPtr<T>(pointer); }
+template<class T> std::vector<T> copy_values(const T* pointer, size_t count) {
+    return pointer ? std::vector<T>(pointer, pointer + count) : std::vector<T>();
+}
+template<class T> T* values_pointer(std::vector<T>& values) { return values.empty() ? nullptr : values.data(); }
+template<class T> struct RetainedArray { std::vector<ComPtr<T>> owners; std::vector<T*> raw; };
+template<class T> RetainedArray<T> retain_values(T* const* pointer, size_t count) {
+    RetainedArray<T> result;
+    for (size_t i = 0; pointer && i < count; ++i) { result.owners.emplace_back(pointer[i]); result.raw.push_back(pointer[i]); }
+    return result;
+}
+struct Hook { void* target; void* replacement; };
+std::vector<Hook> attached;
+void attach_method(void* target, void* replacement, void** original, const char* name) {
+    for (const auto& hook : attached) if (hook.target == target)
+        throw std::runtime_error(std::string("Shared D3D12 implementation needs an explicit alias adapter: ") + name);
+    const auto result = MH_CreateHook(target, replacement, original);
+    if (result != MH_OK) throw std::runtime_error(std::string("Cannot intercept D3D12 ") + name + ": " + MH_StatusToString(result));
+    attached.push_back({target, replacement});
+    std::printf("D4R_D3D12_HOOK method=%s\n", name);
+}
+
+#include "d3d12_command_hooks.generated.h"
+
+using ReleaseFn = ULONG(STDMETHODCALLTYPE*)(IUnknown*);
+ReleaseFn original_release = nullptr;
+ULONG STDMETHODCALLTYPE hook_release(IUnknown* self) {
+    const ULONG count = original_release(self);
+    if (!count) forget(reinterpret_cast<ID3D12GraphicsCommandList*>(self));
+    return count;
+}
+HRESULT STDMETHODCALLTYPE hook_Close(ID3D12GraphicsCommandList* self) {
+    Access access(self);
+    const auto result = original_Close(access.target());
+    if (SUCCEEDED(result) && access.recording) access.recording->closed = true;
+    return result;
+}
+HRESULT STDMETHODCALLTYPE hook_Reset(ID3D12GraphicsCommandList* self, ID3D12CommandAllocator* allocator, ID3D12PipelineState* initial_state) {
+    if (internal) return original_Reset(self, allocator, initial_state);
+    forget(self);
+    const auto result = original_Reset(self, allocator, initial_state);
+    if (SUCCEEDED(result)) {
+        auto recording = get_recording(self);
+        auto pipeline = retain(initial_state);
+        recording->state.emplace_back([pipeline](ID3D12GraphicsCommandList* target) {
+            if (pipeline) original_SetPipelineState(target, pipeline.Get());
+        });
+    }
+    return result;
+}
+void STDMETHODCALLTYPE hook_ClearState(ID3D12GraphicsCommandList* self, ID3D12PipelineState* pipeline_state) {
+    Access access(self);
+    if (access.recording) {
+        access.recording->state.clear(); access.recording->indirect_state_unknown = false;
+        auto pipeline = retain(pipeline_state);
+        access.recording->state.emplace_back([pipeline](ID3D12GraphicsCommandList* target) { original_ClearState(target, pipeline.Get()); });
+    }
+    original_ClearState(access.target(), pipeline_state);
+}
+void STDMETHODCALLTYPE hook_ResourceBarrier(ID3D12GraphicsCommandList* self, UINT barrier_count, const D3D12_RESOURCE_BARRIER* barriers) {
+    Access access(self);
+    if (access.recording) for (UINT i = 0; i < barrier_count; ++i)
+        if (barriers[i].Type == D3D12_RESOURCE_BARRIER_TYPE_TRANSITION && barriers[i].Flags != D3D12_RESOURCE_BARRIER_FLAG_BEGIN_ONLY)
+            access.recording->resource_states[barriers[i].Transition.pResource] = barriers[i].Transition.StateAfter;
+    original_ResourceBarrier(access.target(), barrier_count, barriers);
+}
+void STDMETHODCALLTYPE hook_Barrier(ID3D12GraphicsCommandList* self, UINT32 barrier_groups_count, const D3D12_BARRIER_GROUP* barrier_groups) {
+    Access access(self);
+    original_Barrier(access.target(), barrier_groups_count, barrier_groups);
+    // Enhanced layout-to-legacy conversion is an explicit subsequent gate.
+}
+void STDMETHODCALLTYPE hook_BeginEvent(ID3D12GraphicsCommandList* self, UINT metadata, const void* data, UINT size) {
+    Access access(self);
+    if (access.recording) access.recording->events.push_back({metadata, copy_values(static_cast<const uint8_t*>(data), size)});
+    original_BeginEvent(access.target(), metadata, data, size);
+}
+void STDMETHODCALLTYPE hook_EndEvent(ID3D12GraphicsCommandList* self) {
+    Access access(self);
+    if (access.recording && !access.recording->events.empty()) access.recording->events.pop_back();
+    original_EndEvent(access.target());
+}
+void STDMETHODCALLTYPE hook_BeginRenderPass(ID3D12GraphicsCommandList* self, UINT render_target_count,
+    const D3D12_RENDER_PASS_RENDER_TARGET_DESC* render_targets, const D3D12_RENDER_PASS_DEPTH_STENCIL_DESC* depth_stencil, D3D12_RENDER_PASS_FLAGS flags) {
+    Access access(self); if (access.recording) access.recording->render_pass = true;
+    original_BeginRenderPass(access.target(), render_target_count, render_targets, depth_stencil, flags);
+}
+void STDMETHODCALLTYPE hook_EndRenderPass(ID3D12GraphicsCommandList* self) {
+    Access access(self); if (access.recording) access.recording->render_pass = false;
+    original_EndRenderPass(access.target());
+}
+void STDMETHODCALLTYPE hook_BeginQuery(ID3D12GraphicsCommandList* self, ID3D12QueryHeap* heap, D3D12_QUERY_TYPE type, UINT index) {
+    Access access(self); if (access.recording) ++access.recording->active_queries;
+    original_BeginQuery(access.target(), heap, type, index);
+}
+void STDMETHODCALLTYPE hook_EndQuery(ID3D12GraphicsCommandList* self, ID3D12QueryHeap* heap, D3D12_QUERY_TYPE type, UINT index) {
+    Access access(self); if (access.recording && access.recording->active_queries) --access.recording->active_queries;
+    original_EndQuery(access.target(), heap, type, index);
+}
+void STDMETHODCALLTYPE hook_SetPredication(ID3D12GraphicsCommandList* self, ID3D12Resource* buffer, UINT64 aligned_buffer_offset, D3D12_PREDICATION_OP operation) {
+    Access access(self);
+    if (access.recording) {
+        access.recording->predication = buffer != nullptr;
+        auto retained = retain(buffer);
+        access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) { original_SetPredication(target, retained.Get(), aligned_buffer_offset, operation); });
+    }
+    original_SetPredication(access.target(), buffer, aligned_buffer_offset, operation);
+}
+void STDMETHODCALLTYPE hook_ExecuteBundle(ID3D12GraphicsCommandList* self, ID3D12GraphicsCommandList* command_list) {
+    Access access(self);
+    auto bundle = get_recording(command_list, false);
+    if (access.recording && bundle) {
+        std::lock_guard<std::recursive_mutex> lock(bundle->mutex);
+        access.recording->state.insert(access.recording->state.end(), bundle->state.begin(), bundle->state.end());
+        access.recording->indirect_state_unknown |= bundle->indirect_state_unknown;
+    } else if (access.recording) access.recording->indirect_state_unknown = true;
+    original_ExecuteBundle(access.target(), command_list);
+}
+void STDMETHODCALLTYPE hook_ExecuteIndirect(ID3D12GraphicsCommandList* self, ID3D12CommandSignature* command_signature,
+    UINT max_command_count, ID3D12Resource* arg_buffer, UINT64 arg_buffer_offset, ID3D12Resource* count_buffer, UINT64 count_buffer_offset) {
+    Access access(self);
+    if (access.recording) access.recording->indirect_state_unknown = true;
+    original_ExecuteIndirect(access.target(), command_signature, max_command_count, arg_buffer, arg_buffer_offset, count_buffer, count_buffer_offset);
+}
+
+using ExecuteFn = void(STDMETHODCALLTYPE*)(ID3D12CommandQueue*, UINT, ID3D12CommandList* const*);
+ExecuteFn original_execute = nullptr;
+void complete(ID3D12CommandQueue* queue) {
+    ComPtr<ID3D12Device> device; dx(queue->GetDevice(IID_PPV_ARGS(device.GetAddressOf())), "Queue device(submit drain)");
+    ComPtr<ID3D12Fence> fence; dx(device->CreateFence(0, D3D12_FENCE_FLAG_NONE, IID_PPV_ARGS(fence.GetAddressOf())), "Queue fence(submit drain)");
+    Handle event; event.value = CreateEventW(nullptr, FALSE, FALSE, nullptr);
+    if (!event.value) throw std::runtime_error("CreateEvent(submit drain)");
+    dx(queue->Signal(fence.Get(), 1), "Queue Signal(submit drain)");
+    if (fence->GetCompletedValue() < 1) {
+        dx(fence->SetEventOnCompletion(1, event.value), "Queue event(submit drain)");
+        if (WaitForSingleObject(event.value, 30000) != WAIT_OBJECT_0) throw std::runtime_error("GPU submit boundary timed out");
+    }
+}
+void STDMETHODCALLTYPE hook_execute(ID3D12CommandQueue* queue, UINT count, ID3D12CommandList* const* lists) {
+    if (internal) { original_execute(queue, count, lists); return; }
+    std::vector<std::shared_ptr<Recording>> owners;
+    std::vector<ID3D12CommandList*> pending;
+    InternalScope scope;
+    auto flush = [&] { if (!pending.empty()) { original_execute(queue, UINT(pending.size()), pending.data()); pending.clear(); } };
+    bool boundaries = false;
+    try {
+        for (UINT i = 0; i < count; ++i) {
+            std::shared_ptr<Recording> recording;
+            { std::lock_guard<std::mutex> lock(map_mutex);
+              auto found = recordings.find(static_cast<ID3D12GraphicsCommandList*>(lists[i]));
+              if (found != recordings.end()) recording = found->second; }
+            if (!recording || recording->segments.empty()) { pending.push_back(lists[i]); continue; }
+            std::lock_guard<std::recursive_mutex> lock(recording->mutex);
+            if (!recording->closed) throw std::runtime_error("Logical command list submitted before Close");
+            owners.push_back(recording); boundaries = true;
+            for (const auto& segment : recording->segments) {
+                pending.push_back(segment.list); flush();
+                segment.boundary(queue);
+            }
+            pending.push_back(recording->current.Get());
+        }
+        flush();
+        if (boundaries) complete(queue);
+    } catch (const std::exception& failure) {
+        { std::lock_guard<std::mutex> lock(error_mutex); last_error = failure.what(); }
+        std::fprintf(stderr, "D4R_D3D12_SUBMIT_FAILURE %s\n", failure.what());
+        try { complete(queue); } catch (...) {}
+        // Do not submit a consumer suffix after a failed DLSS boundary.
+    }
+}
+}
+
+InternalScope::InternalScope() : previous(internal) { internal = true; }
+InternalScope::~InternalScope() { internal = previous; }
+std::string error() { std::lock_guard<std::mutex> lock(error_mutex); return last_error; }
+size_t live_recordings() { std::lock_guard<std::mutex> lock(map_mutex); return recordings.size(); }
+D3D12_RESOURCE_STATES resource_state(ID3D12GraphicsCommandList* list, ID3D12Resource* resource, D3D12_RESOURCE_STATES fallback) {
+    auto recording = get_recording(list, false);
+    if (!recording) return fallback;
+    std::lock_guard<std::recursive_mutex> lock(recording->mutex);
+    auto found = recording->resource_states.find(resource);
+    return found == recording->resource_states.end() ? fallback : found->second;
+}
+void install(ID3D12Device* device) {
+    std::lock_guard<std::mutex> lock(installation_mutex);
+    if (installed) return;
+    InternalScope scope;
+    // An application may unload its NGX handle while native D3D12 objects are
+    // still alive. Public method callbacks must remain callable until exit.
+    HMODULE owner = nullptr;
+    if (!GetModuleHandleExW(GET_MODULE_HANDLE_EX_FLAG_FROM_ADDRESS | GET_MODULE_HANDLE_EX_FLAG_PIN,
+        reinterpret_cast<LPCWSTR>(&install), &owner)) throw std::runtime_error("Cannot retain the D3D12 callback module");
+    const auto init = MH_Initialize();
+    if (init != MH_OK && init != MH_ERROR_ALREADY_INITIALIZED) throw std::runtime_error(MH_StatusToString(init));
+    ComPtr<ID3D12CommandAllocator> allocator; ComPtr<ID3D12GraphicsCommandList> list;
+    dx(device->CreateCommandAllocator(D3D12_COMMAND_LIST_TYPE_DIRECT, IID_PPV_ARGS(allocator.GetAddressOf())), "Create hook allocator");
+    dx(device->CreateCommandList(0, D3D12_COMMAND_LIST_TYPE_DIRECT, allocator.Get(), nullptr, IID_PPV_ARGS(list.GetAddressOf())), "Create hook command list");
+    try {
+        // The public COM interface inherits one ordered method table.
+        unsigned version = 0;
+        ComPtr<IUnknown> newest;
+        const GUID* ids[] = {&IID_ID3D12GraphicsCommandList, &IID_ID3D12GraphicsCommandList1, &IID_ID3D12GraphicsCommandList2,
+            &IID_ID3D12GraphicsCommandList3, &IID_ID3D12GraphicsCommandList4, &IID_ID3D12GraphicsCommandList5,
+            &IID_ID3D12GraphicsCommandList6, &IID_ID3D12GraphicsCommandList7};
+        for (unsigned n = 0; n < 8; ++n) if (SUCCEEDED(list->QueryInterface(*ids[n], reinterpret_cast<void**>(newest.ReleaseAndGetAddressOf())))) version = n;
+        auto table = *reinterpret_cast<void***>(list.Get());
+        attach_recording_methods(table, version);
+        attach_method(table[2], reinterpret_cast<void*>(hook_release), reinterpret_cast<void**>(&original_release), "Release");
+        D3D12_COMMAND_QUEUE_DESC desc{}; desc.Type = D3D12_COMMAND_LIST_TYPE_DIRECT;
+        ComPtr<ID3D12CommandQueue> queue; dx(device->CreateCommandQueue(&desc, IID_PPV_ARGS(queue.GetAddressOf())), "Create hook queue");
+        auto queueTable = *reinterpret_cast<void***>(queue.Get());
+        // Queue inherits IUnknown, Object, DeviceChild; UpdateTileMappings,
+        // CopyTileMappings precede ExecuteCommandLists in the public SDK.
+        attach_method(queueTable[10], reinterpret_cast<void*>(hook_execute), reinterpret_cast<void**>(&original_execute), "ExecuteCommandLists");
+        for (const auto& hook : attached) {
+            const auto result = MH_QueueEnableHook(hook.target);
+            if (result != MH_OK) throw std::runtime_error(MH_StatusToString(result));
+        }
+        const auto result = MH_ApplyQueued();
+        if (result != MH_OK) throw std::runtime_error(MH_StatusToString(result));
+        installed = true;
+        std::printf("D4R_D3D12_COMMAND_BACKEND installed=1 interface=%u hooks=%zu frame_age=0\n", version, attached.size());
+    } catch (...) {
+        for (const auto& hook : attached) { (void)MH_DisableHook(hook.target); (void)MH_RemoveHook(hook.target); }
+        attached.clear(); throw;
+    }
+}
+void record_boundary(ID3D12GraphicsCommandList* list, Boundary callback) {
+    if (!installed || !list || !callback) throw std::runtime_error("D3D12 command backend is not initialized");
+    if (!error().empty()) throw std::runtime_error(error());
+    if (list->GetType() != D3D12_COMMAND_LIST_TYPE_DIRECT) throw std::runtime_error("DLSS requires a direct command list");
+    Access access(list);
+    auto& recording = *access.recording;
+    if (recording.closed || recording.render_pass || recording.active_queries || recording.predication || recording.indirect_state_unknown)
+        throw std::runtime_error("Cannot split a closed list or an active render/query/predication/unknown indirect state scope");
+    // Construct/replay the suffix before modifying the original list so an
+    // allocation failure leaves the caller's recording intact.
+    InternalScope scope;
+    ComPtr<ID3D12Device> device; dx(list->GetDevice(IID_PPV_ARGS(device.GetAddressOf())), "Boundary device");
+    Segment segment; segment.list = access.target(); segment.owner = recording.current;
+    segment.allocator = recording.current_allocator; segment.boundary = std::move(callback);
+    ComPtr<ID3D12CommandAllocator> nextAllocator; ComPtr<ID3D12GraphicsCommandList> next;
+    dx(device->CreateCommandAllocator(D3D12_COMMAND_LIST_TYPE_DIRECT, IID_PPV_ARGS(nextAllocator.GetAddressOf())), "Suffix allocator");
+    dx(device->CreateCommandList(0, D3D12_COMMAND_LIST_TYPE_DIRECT, nextAllocator.Get(), nullptr, IID_PPV_ARGS(next.GetAddressOf())), "Suffix command list");
+    for (auto& state : recording.state) state(next.Get());
+    recording.segments.reserve(recording.segments.size() + 1);
+    for (const auto& event : recording.events) original_BeginEvent(next.Get(), event.metadata, event.data.data(), UINT(event.data.size()));
+    for (size_t n = 0; n < recording.events.size(); ++n) original_EndEvent(segment.list);
+    dx(original_Close(segment.list), "Close boundary prefix");
+    // Each suffix owns an independent allocator until GPU completion/reset.
+    recording.segments.push_back(std::move(segment));
+    recording.current_allocator = std::move(nextAllocator);
+    recording.current = std::move(next);
+    std::printf("D4R_D3D12_RECORD boundary=%zu state_calls=%zu frame_age=0\n", recording.segments.size(), recording.state.size());
+}
+}

tools/windows/d3d12_command_hooks.generated.h

@@ -0,0 +1,653 @@
+// Generated by generate_d3d12_hooks.py from the public SDK. Do not edit.
+
+using Fn_Close = HRESULT(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self);
+static Fn_Close original_Close = nullptr;
+static HRESULT STDMETHODCALLTYPE hook_Close(ID3D12GraphicsCommandList* self);
+
+using Fn_Reset = HRESULT(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12CommandAllocator *allocator, ID3D12PipelineState *initial_state);
+static Fn_Reset original_Reset = nullptr;
+static HRESULT STDMETHODCALLTYPE hook_Reset(ID3D12GraphicsCommandList* self, ID3D12CommandAllocator *allocator, ID3D12PipelineState *initial_state);
+
+using Fn_ClearState = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12PipelineState *pipeline_state);
+static Fn_ClearState original_ClearState = nullptr;
+static void STDMETHODCALLTYPE hook_ClearState(ID3D12GraphicsCommandList* self, ID3D12PipelineState *pipeline_state);
+
+using Fn_DrawInstanced = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT vertex_count_per_instance, UINT instance_count, UINT start_vertex_location, UINT start_instance_location);
+static Fn_DrawInstanced original_DrawInstanced = nullptr;
+static void STDMETHODCALLTYPE hook_DrawInstanced(ID3D12GraphicsCommandList* self, UINT vertex_count_per_instance, UINT instance_count, UINT start_vertex_location, UINT start_instance_location) {
+    Access access(self);
+    return original_DrawInstanced(access.target(), vertex_count_per_instance, instance_count, start_vertex_location, start_instance_location);
+}
+
+using Fn_DrawIndexedInstanced = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT index_count_per_instance, UINT instance_count, UINT start_vertex_location, INT base_vertex_location, UINT start_instance_location);
+static Fn_DrawIndexedInstanced original_DrawIndexedInstanced = nullptr;
+static void STDMETHODCALLTYPE hook_DrawIndexedInstanced(ID3D12GraphicsCommandList* self, UINT index_count_per_instance, UINT instance_count, UINT start_vertex_location, INT base_vertex_location, UINT start_instance_location) {
+    Access access(self);
+    return original_DrawIndexedInstanced(access.target(), index_count_per_instance, instance_count, start_vertex_location, base_vertex_location, start_instance_location);
+}
+
+using Fn_Dispatch = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT x, UINT u, UINT z);
+static Fn_Dispatch original_Dispatch = nullptr;
+static void STDMETHODCALLTYPE hook_Dispatch(ID3D12GraphicsCommandList* self, UINT x, UINT u, UINT z) {
+    Access access(self);
+    return original_Dispatch(access.target(), x, u, z);
+}
+
+using Fn_CopyBufferRegion = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_buffer, UINT64 dst_offset, ID3D12Resource *src_buffer, UINT64 src_offset, UINT64 byte_count);
+static Fn_CopyBufferRegion original_CopyBufferRegion = nullptr;
+static void STDMETHODCALLTYPE hook_CopyBufferRegion(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_buffer, UINT64 dst_offset, ID3D12Resource *src_buffer, UINT64 src_offset, UINT64 byte_count) {
+    Access access(self);
+    return original_CopyBufferRegion(access.target(), dst_buffer, dst_offset, src_buffer, src_offset, byte_count);
+}
+
+using Fn_CopyTextureRegion = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, const D3D12_TEXTURE_COPY_LOCATION *dst, UINT dst_x, UINT dst_y, UINT dst_z, const D3D12_TEXTURE_COPY_LOCATION *src, const D3D12_BOX *src_box);
+static Fn_CopyTextureRegion original_CopyTextureRegion = nullptr;
+static void STDMETHODCALLTYPE hook_CopyTextureRegion(ID3D12GraphicsCommandList* self, const D3D12_TEXTURE_COPY_LOCATION *dst, UINT dst_x, UINT dst_y, UINT dst_z, const D3D12_TEXTURE_COPY_LOCATION *src, const D3D12_BOX *src_box) {
+    Access access(self);
+    return original_CopyTextureRegion(access.target(), dst, dst_x, dst_y, dst_z, src, src_box);
+}
+
+using Fn_CopyResource = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_resource, ID3D12Resource *src_resource);
+static Fn_CopyResource original_CopyResource = nullptr;
+static void STDMETHODCALLTYPE hook_CopyResource(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_resource, ID3D12Resource *src_resource) {
+    Access access(self);
+    return original_CopyResource(access.target(), dst_resource, src_resource);
+}
+
+using Fn_CopyTiles = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12Resource *tiled_resource, const D3D12_TILED_RESOURCE_COORDINATE *tile_region_start_coordinate, const D3D12_TILE_REGION_SIZE *tile_region_size, ID3D12Resource *buffer, UINT64 buffer_offset, D3D12_TILE_COPY_FLAGS flags);
+static Fn_CopyTiles original_CopyTiles = nullptr;
+static void STDMETHODCALLTYPE hook_CopyTiles(ID3D12GraphicsCommandList* self, ID3D12Resource *tiled_resource, const D3D12_TILED_RESOURCE_COORDINATE *tile_region_start_coordinate, const D3D12_TILE_REGION_SIZE *tile_region_size, ID3D12Resource *buffer, UINT64 buffer_offset, D3D12_TILE_COPY_FLAGS flags) {
+    Access access(self);
+    return original_CopyTiles(access.target(), tiled_resource, tile_region_start_coordinate, tile_region_size, buffer, buffer_offset, flags);
+}
+
+using Fn_ResolveSubresource = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_resource, UINT dst_sub_resource, ID3D12Resource *src_resource, UINT src_sub_resource, DXGI_FORMAT format);
+static Fn_ResolveSubresource original_ResolveSubresource = nullptr;
+static void STDMETHODCALLTYPE hook_ResolveSubresource(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_resource, UINT dst_sub_resource, ID3D12Resource *src_resource, UINT src_sub_resource, DXGI_FORMAT format) {
+    Access access(self);
+    return original_ResolveSubresource(access.target(), dst_resource, dst_sub_resource, src_resource, src_sub_resource, format);
+}
+
+using Fn_IASetPrimitiveTopology = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, D3D12_PRIMITIVE_TOPOLOGY primitive_topology);
+static Fn_IASetPrimitiveTopology original_IASetPrimitiveTopology = nullptr;
+static void STDMETHODCALLTYPE hook_IASetPrimitiveTopology(ID3D12GraphicsCommandList* self, D3D12_PRIMITIVE_TOPOLOGY primitive_topology) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_IASetPrimitiveTopology(target, primitive_topology);
+    });
+    return original_IASetPrimitiveTopology(access.target(), primitive_topology);
+}
+
+using Fn_RSSetViewports = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT viewport_count, const D3D12_VIEWPORT *viewports);
+static Fn_RSSetViewports original_RSSetViewports = nullptr;
+static void STDMETHODCALLTYPE hook_RSSetViewports(ID3D12GraphicsCommandList* self, UINT viewport_count, const D3D12_VIEWPORT *viewports) {
+    Access access(self);
+    auto saved_viewports = copy_values(viewports, viewport_count);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_RSSetViewports(target, viewport_count, values_pointer(saved_viewports));
+    });
+    return original_RSSetViewports(access.target(), viewport_count, viewports);
+}
+
+using Fn_RSSetScissorRects = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT rect_count, const D3D12_RECT *rects);
+static Fn_RSSetScissorRects original_RSSetScissorRects = nullptr;
+static void STDMETHODCALLTYPE hook_RSSetScissorRects(ID3D12GraphicsCommandList* self, UINT rect_count, const D3D12_RECT *rects) {
+    Access access(self);
+    auto saved_rects = copy_values(rects, rect_count);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_RSSetScissorRects(target, rect_count, values_pointer(saved_rects));
+    });
+    return original_RSSetScissorRects(access.target(), rect_count, rects);
+}
+
+using Fn_OMSetBlendFactor = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, const FLOAT blend_factor[4]);
+static Fn_OMSetBlendFactor original_OMSetBlendFactor = nullptr;
+static void STDMETHODCALLTYPE hook_OMSetBlendFactor(ID3D12GraphicsCommandList* self, const FLOAT blend_factor[4]) {
+    Access access(self);
+    auto saved_blend_factor = copy_values(blend_factor, 4);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_OMSetBlendFactor(target, values_pointer(saved_blend_factor));
+    });
+    return original_OMSetBlendFactor(access.target(), blend_factor);
+}
+
+using Fn_OMSetStencilRef = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT stencil_ref);
+static Fn_OMSetStencilRef original_OMSetStencilRef = nullptr;
+static void STDMETHODCALLTYPE hook_OMSetStencilRef(ID3D12GraphicsCommandList* self, UINT stencil_ref) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_OMSetStencilRef(target, stencil_ref);
+    });
+    return original_OMSetStencilRef(access.target(), stencil_ref);
+}
+
+using Fn_SetPipelineState = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12PipelineState *pipeline_state);
+static Fn_SetPipelineState original_SetPipelineState = nullptr;
+static void STDMETHODCALLTYPE hook_SetPipelineState(ID3D12GraphicsCommandList* self, ID3D12PipelineState *pipeline_state) {
+    Access access(self);
+    auto saved_pipeline_state = retain(pipeline_state);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetPipelineState(target, saved_pipeline_state.Get());
+    });
+    return original_SetPipelineState(access.target(), pipeline_state);
+}
+
+using Fn_ResourceBarrier = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT barrier_count, const D3D12_RESOURCE_BARRIER *barriers);
+static Fn_ResourceBarrier original_ResourceBarrier = nullptr;
+static void STDMETHODCALLTYPE hook_ResourceBarrier(ID3D12GraphicsCommandList* self, UINT barrier_count, const D3D12_RESOURCE_BARRIER *barriers);
+
+using Fn_ExecuteBundle = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12GraphicsCommandList *command_list);
+static Fn_ExecuteBundle original_ExecuteBundle = nullptr;
+static void STDMETHODCALLTYPE hook_ExecuteBundle(ID3D12GraphicsCommandList* self, ID3D12GraphicsCommandList *command_list);
+
+using Fn_SetDescriptorHeaps = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT heap_count, ID3D12DescriptorHeap *const *heaps);
+static Fn_SetDescriptorHeaps original_SetDescriptorHeaps = nullptr;
+static void STDMETHODCALLTYPE hook_SetDescriptorHeaps(ID3D12GraphicsCommandList* self, UINT heap_count, ID3D12DescriptorHeap *const *heaps) {
+    Access access(self);
+    auto saved_heaps = retain_values(heaps, heap_count);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetDescriptorHeaps(target, heap_count, saved_heaps.raw.data());
+    });
+    return original_SetDescriptorHeaps(access.target(), heap_count, heaps);
+}
+
+using Fn_SetComputeRootSignature = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12RootSignature *root_signature);
+static Fn_SetComputeRootSignature original_SetComputeRootSignature = nullptr;
+static void STDMETHODCALLTYPE hook_SetComputeRootSignature(ID3D12GraphicsCommandList* self, ID3D12RootSignature *root_signature) {
+    Access access(self);
+    auto saved_root_signature = retain(root_signature);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetComputeRootSignature(target, saved_root_signature.Get());
+    });
+    return original_SetComputeRootSignature(access.target(), root_signature);
+}
+
+using Fn_SetGraphicsRootSignature = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12RootSignature *root_signature);
+static Fn_SetGraphicsRootSignature original_SetGraphicsRootSignature = nullptr;
+static void STDMETHODCALLTYPE hook_SetGraphicsRootSignature(ID3D12GraphicsCommandList* self, ID3D12RootSignature *root_signature) {
+    Access access(self);
+    auto saved_root_signature = retain(root_signature);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetGraphicsRootSignature(target, saved_root_signature.Get());
+    });
+    return original_SetGraphicsRootSignature(access.target(), root_signature);
+}
+
+using Fn_SetComputeRootDescriptorTable = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_DESCRIPTOR_HANDLE base_descriptor);
+static Fn_SetComputeRootDescriptorTable original_SetComputeRootDescriptorTable = nullptr;
+static void STDMETHODCALLTYPE hook_SetComputeRootDescriptorTable(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_DESCRIPTOR_HANDLE base_descriptor) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetComputeRootDescriptorTable(target, root_parameter_index, base_descriptor);
+    });
+    return original_SetComputeRootDescriptorTable(access.target(), root_parameter_index, base_descriptor);
+}
+
+using Fn_SetGraphicsRootDescriptorTable = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_DESCRIPTOR_HANDLE base_descriptor);
+static Fn_SetGraphicsRootDescriptorTable original_SetGraphicsRootDescriptorTable = nullptr;
+static void STDMETHODCALLTYPE hook_SetGraphicsRootDescriptorTable(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_DESCRIPTOR_HANDLE base_descriptor) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetGraphicsRootDescriptorTable(target, root_parameter_index, base_descriptor);
+    });
+    return original_SetGraphicsRootDescriptorTable(access.target(), root_parameter_index, base_descriptor);
+}
+
+using Fn_SetComputeRoot32BitConstant = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, UINT data, UINT dst_offset);
+static Fn_SetComputeRoot32BitConstant original_SetComputeRoot32BitConstant = nullptr;
+static void STDMETHODCALLTYPE hook_SetComputeRoot32BitConstant(ID3D12GraphicsCommandList* self, UINT root_parameter_index, UINT data, UINT dst_offset) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetComputeRoot32BitConstant(target, root_parameter_index, data, dst_offset);
+    });
+    return original_SetComputeRoot32BitConstant(access.target(), root_parameter_index, data, dst_offset);
+}
+
+using Fn_SetGraphicsRoot32BitConstant = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, UINT data, UINT dst_offset);
+static Fn_SetGraphicsRoot32BitConstant original_SetGraphicsRoot32BitConstant = nullptr;
+static void STDMETHODCALLTYPE hook_SetGraphicsRoot32BitConstant(ID3D12GraphicsCommandList* self, UINT root_parameter_index, UINT data, UINT dst_offset) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetGraphicsRoot32BitConstant(target, root_parameter_index, data, dst_offset);
+    });
+    return original_SetGraphicsRoot32BitConstant(access.target(), root_parameter_index, data, dst_offset);
+}
+
+using Fn_SetComputeRoot32BitConstants = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, UINT constant_count, const void *data, UINT dst_offset);
+static Fn_SetComputeRoot32BitConstants original_SetComputeRoot32BitConstants = nullptr;
+static void STDMETHODCALLTYPE hook_SetComputeRoot32BitConstants(ID3D12GraphicsCommandList* self, UINT root_parameter_index, UINT constant_count, const void *data, UINT dst_offset) {
+    Access access(self);
+    auto saved_data = copy_values(static_cast<const UINT*>(data), constant_count);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetComputeRoot32BitConstants(target, root_parameter_index, constant_count, values_pointer(saved_data), dst_offset);
+    });
+    return original_SetComputeRoot32BitConstants(access.target(), root_parameter_index, constant_count, data, dst_offset);
+}
+
+using Fn_SetGraphicsRoot32BitConstants = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, UINT constant_count, const void *data, UINT dst_offset);
+static Fn_SetGraphicsRoot32BitConstants original_SetGraphicsRoot32BitConstants = nullptr;
+static void STDMETHODCALLTYPE hook_SetGraphicsRoot32BitConstants(ID3D12GraphicsCommandList* self, UINT root_parameter_index, UINT constant_count, const void *data, UINT dst_offset) {
+    Access access(self);
+    auto saved_data = copy_values(static_cast<const UINT*>(data), constant_count);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetGraphicsRoot32BitConstants(target, root_parameter_index, constant_count, values_pointer(saved_data), dst_offset);
+    });
+    return original_SetGraphicsRoot32BitConstants(access.target(), root_parameter_index, constant_count, data, dst_offset);
+}
+
+using Fn_SetComputeRootConstantBufferView = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address);
+static Fn_SetComputeRootConstantBufferView original_SetComputeRootConstantBufferView = nullptr;
+static void STDMETHODCALLTYPE hook_SetComputeRootConstantBufferView(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetComputeRootConstantBufferView(target, root_parameter_index, address);
+    });
+    return original_SetComputeRootConstantBufferView(access.target(), root_parameter_index, address);
+}
+
+using Fn_SetGraphicsRootConstantBufferView = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address);
+static Fn_SetGraphicsRootConstantBufferView original_SetGraphicsRootConstantBufferView = nullptr;
+static void STDMETHODCALLTYPE hook_SetGraphicsRootConstantBufferView(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetGraphicsRootConstantBufferView(target, root_parameter_index, address);
+    });
+    return original_SetGraphicsRootConstantBufferView(access.target(), root_parameter_index, address);
+}
+
+using Fn_SetComputeRootShaderResourceView = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address);
+static Fn_SetComputeRootShaderResourceView original_SetComputeRootShaderResourceView = nullptr;
+static void STDMETHODCALLTYPE hook_SetComputeRootShaderResourceView(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetComputeRootShaderResourceView(target, root_parameter_index, address);
+    });
+    return original_SetComputeRootShaderResourceView(access.target(), root_parameter_index, address);
+}
+
+using Fn_SetGraphicsRootShaderResourceView = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address);
+static Fn_SetGraphicsRootShaderResourceView original_SetGraphicsRootShaderResourceView = nullptr;
+static void STDMETHODCALLTYPE hook_SetGraphicsRootShaderResourceView(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetGraphicsRootShaderResourceView(target, root_parameter_index, address);
+    });
+    return original_SetGraphicsRootShaderResourceView(access.target(), root_parameter_index, address);
+}
+
+using Fn_SetComputeRootUnorderedAccessView = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address);
+static Fn_SetComputeRootUnorderedAccessView original_SetComputeRootUnorderedAccessView = nullptr;
+static void STDMETHODCALLTYPE hook_SetComputeRootUnorderedAccessView(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetComputeRootUnorderedAccessView(target, root_parameter_index, address);
+    });
+    return original_SetComputeRootUnorderedAccessView(access.target(), root_parameter_index, address);
+}
+
+using Fn_SetGraphicsRootUnorderedAccessView = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address);
+static Fn_SetGraphicsRootUnorderedAccessView original_SetGraphicsRootUnorderedAccessView = nullptr;
+static void STDMETHODCALLTYPE hook_SetGraphicsRootUnorderedAccessView(ID3D12GraphicsCommandList* self, UINT root_parameter_index, D3D12_GPU_VIRTUAL_ADDRESS address) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetGraphicsRootUnorderedAccessView(target, root_parameter_index, address);
+    });
+    return original_SetGraphicsRootUnorderedAccessView(access.target(), root_parameter_index, address);
+}
+
+using Fn_IASetIndexBuffer = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, const D3D12_INDEX_BUFFER_VIEW *view);
+static Fn_IASetIndexBuffer original_IASetIndexBuffer = nullptr;
+static void STDMETHODCALLTYPE hook_IASetIndexBuffer(ID3D12GraphicsCommandList* self, const D3D12_INDEX_BUFFER_VIEW *view) {
+    Access access(self);
+    auto saved_view = copy_values(view, 1);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_IASetIndexBuffer(target, values_pointer(saved_view));
+    });
+    return original_IASetIndexBuffer(access.target(), view);
+}
+
+using Fn_IASetVertexBuffers = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT start_slot, UINT view_count, const D3D12_VERTEX_BUFFER_VIEW *views);
+static Fn_IASetVertexBuffers original_IASetVertexBuffers = nullptr;
+static void STDMETHODCALLTYPE hook_IASetVertexBuffers(ID3D12GraphicsCommandList* self, UINT start_slot, UINT view_count, const D3D12_VERTEX_BUFFER_VIEW *views) {
+    Access access(self);
+    auto saved_views = copy_values(views, view_count);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_IASetVertexBuffers(target, start_slot, view_count, values_pointer(saved_views));
+    });
+    return original_IASetVertexBuffers(access.target(), start_slot, view_count, views);
+}
+
+using Fn_SOSetTargets = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT start_slot, UINT view_count, const D3D12_STREAM_OUTPUT_BUFFER_VIEW *views);
+static Fn_SOSetTargets original_SOSetTargets = nullptr;
+static void STDMETHODCALLTYPE hook_SOSetTargets(ID3D12GraphicsCommandList* self, UINT start_slot, UINT view_count, const D3D12_STREAM_OUTPUT_BUFFER_VIEW *views) {
+    Access access(self);
+    auto saved_views = copy_values(views, view_count);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SOSetTargets(target, start_slot, view_count, values_pointer(saved_views));
+    });
+    return original_SOSetTargets(access.target(), start_slot, view_count, views);
+}
+
+using Fn_OMSetRenderTargets = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT render_target_descriptor_count, const D3D12_CPU_DESCRIPTOR_HANDLE *render_target_descriptors, WINBOOL single_descriptor_handle, const D3D12_CPU_DESCRIPTOR_HANDLE *depth_stencil_descriptor);
+static Fn_OMSetRenderTargets original_OMSetRenderTargets = nullptr;
+static void STDMETHODCALLTYPE hook_OMSetRenderTargets(ID3D12GraphicsCommandList* self, UINT render_target_descriptor_count, const D3D12_CPU_DESCRIPTOR_HANDLE *render_target_descriptors, WINBOOL single_descriptor_handle, const D3D12_CPU_DESCRIPTOR_HANDLE *depth_stencil_descriptor) {
+    Access access(self);
+    auto saved_render_target_descriptors = copy_values(render_target_descriptors, (single_descriptor_handle ? 1 : render_target_descriptor_count));
+    auto saved_depth_stencil_descriptor = copy_values(depth_stencil_descriptor, 1);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_OMSetRenderTargets(target, render_target_descriptor_count, values_pointer(saved_render_target_descriptors), single_descriptor_handle, values_pointer(saved_depth_stencil_descriptor));
+    });
+    return original_OMSetRenderTargets(access.target(), render_target_descriptor_count, render_target_descriptors, single_descriptor_handle, depth_stencil_descriptor);
+}
+
+using Fn_ClearDepthStencilView = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, D3D12_CPU_DESCRIPTOR_HANDLE dsv, D3D12_CLEAR_FLAGS flags, FLOAT depth, UINT8 stencil, UINT rect_count, const D3D12_RECT *rects);
+static Fn_ClearDepthStencilView original_ClearDepthStencilView = nullptr;
+static void STDMETHODCALLTYPE hook_ClearDepthStencilView(ID3D12GraphicsCommandList* self, D3D12_CPU_DESCRIPTOR_HANDLE dsv, D3D12_CLEAR_FLAGS flags, FLOAT depth, UINT8 stencil, UINT rect_count, const D3D12_RECT *rects) {
+    Access access(self);
+    return original_ClearDepthStencilView(access.target(), dsv, flags, depth, stencil, rect_count, rects);
+}
+
+using Fn_ClearRenderTargetView = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, D3D12_CPU_DESCRIPTOR_HANDLE rtv, const FLOAT color[4], UINT rect_count, const D3D12_RECT *rects);
+static Fn_ClearRenderTargetView original_ClearRenderTargetView = nullptr;
+static void STDMETHODCALLTYPE hook_ClearRenderTargetView(ID3D12GraphicsCommandList* self, D3D12_CPU_DESCRIPTOR_HANDLE rtv, const FLOAT color[4], UINT rect_count, const D3D12_RECT *rects) {
+    Access access(self);
+    return original_ClearRenderTargetView(access.target(), rtv, color, rect_count, rects);
+}
+
+using Fn_ClearUnorderedAccessViewUint = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle, D3D12_CPU_DESCRIPTOR_HANDLE cpu_handle, ID3D12Resource *resource, const UINT values[4], UINT rect_count, const D3D12_RECT *rects);
+static Fn_ClearUnorderedAccessViewUint original_ClearUnorderedAccessViewUint = nullptr;
+static void STDMETHODCALLTYPE hook_ClearUnorderedAccessViewUint(ID3D12GraphicsCommandList* self, D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle, D3D12_CPU_DESCRIPTOR_HANDLE cpu_handle, ID3D12Resource *resource, const UINT values[4], UINT rect_count, const D3D12_RECT *rects) {
+    Access access(self);
+    return original_ClearUnorderedAccessViewUint(access.target(), gpu_handle, cpu_handle, resource, values, rect_count, rects);
+}
+
+using Fn_ClearUnorderedAccessViewFloat = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle, D3D12_CPU_DESCRIPTOR_HANDLE cpu_handle, ID3D12Resource *resource, const float values[4], UINT rect_count, const D3D12_RECT *rects);
+static Fn_ClearUnorderedAccessViewFloat original_ClearUnorderedAccessViewFloat = nullptr;
+static void STDMETHODCALLTYPE hook_ClearUnorderedAccessViewFloat(ID3D12GraphicsCommandList* self, D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle, D3D12_CPU_DESCRIPTOR_HANDLE cpu_handle, ID3D12Resource *resource, const float values[4], UINT rect_count, const D3D12_RECT *rects) {
+    Access access(self);
+    return original_ClearUnorderedAccessViewFloat(access.target(), gpu_handle, cpu_handle, resource, values, rect_count, rects);
+}
+
+using Fn_DiscardResource = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12Resource *resource, const D3D12_DISCARD_REGION *region);
+static Fn_DiscardResource original_DiscardResource = nullptr;
+static void STDMETHODCALLTYPE hook_DiscardResource(ID3D12GraphicsCommandList* self, ID3D12Resource *resource, const D3D12_DISCARD_REGION *region) {
+    Access access(self);
+    return original_DiscardResource(access.target(), resource, region);
+}
+
+using Fn_BeginQuery = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12QueryHeap *heap, D3D12_QUERY_TYPE type, UINT index);
+static Fn_BeginQuery original_BeginQuery = nullptr;
+static void STDMETHODCALLTYPE hook_BeginQuery(ID3D12GraphicsCommandList* self, ID3D12QueryHeap *heap, D3D12_QUERY_TYPE type, UINT index);
+
+using Fn_EndQuery = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12QueryHeap *heap, D3D12_QUERY_TYPE type, UINT index);
+static Fn_EndQuery original_EndQuery = nullptr;
+static void STDMETHODCALLTYPE hook_EndQuery(ID3D12GraphicsCommandList* self, ID3D12QueryHeap *heap, D3D12_QUERY_TYPE type, UINT index);
+
+using Fn_ResolveQueryData = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12QueryHeap *heap, D3D12_QUERY_TYPE type, UINT start_index, UINT query_count, ID3D12Resource *dst_buffer, UINT64 aligned_dst_buffer_offset);
+static Fn_ResolveQueryData original_ResolveQueryData = nullptr;
+static void STDMETHODCALLTYPE hook_ResolveQueryData(ID3D12GraphicsCommandList* self, ID3D12QueryHeap *heap, D3D12_QUERY_TYPE type, UINT start_index, UINT query_count, ID3D12Resource *dst_buffer, UINT64 aligned_dst_buffer_offset) {
+    Access access(self);
+    return original_ResolveQueryData(access.target(), heap, type, start_index, query_count, dst_buffer, aligned_dst_buffer_offset);
+}
+
+using Fn_SetPredication = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12Resource *buffer, UINT64 aligned_buffer_offset, D3D12_PREDICATION_OP operation);
+static Fn_SetPredication original_SetPredication = nullptr;
+static void STDMETHODCALLTYPE hook_SetPredication(ID3D12GraphicsCommandList* self, ID3D12Resource *buffer, UINT64 aligned_buffer_offset, D3D12_PREDICATION_OP operation);
+
+using Fn_SetMarker = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT metadata, const void *data, UINT size);
+static Fn_SetMarker original_SetMarker = nullptr;
+static void STDMETHODCALLTYPE hook_SetMarker(ID3D12GraphicsCommandList* self, UINT metadata, const void *data, UINT size) {
+    Access access(self);
+    return original_SetMarker(access.target(), metadata, data, size);
+}
+
+using Fn_BeginEvent = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT metadata, const void *data, UINT size);
+static Fn_BeginEvent original_BeginEvent = nullptr;
+static void STDMETHODCALLTYPE hook_BeginEvent(ID3D12GraphicsCommandList* self, UINT metadata, const void *data, UINT size);
+
+using Fn_EndEvent = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self);
+static Fn_EndEvent original_EndEvent = nullptr;
+static void STDMETHODCALLTYPE hook_EndEvent(ID3D12GraphicsCommandList* self);
+
+using Fn_ExecuteIndirect = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12CommandSignature *command_signature, UINT max_command_count, ID3D12Resource *arg_buffer, UINT64 arg_buffer_offset, ID3D12Resource *count_buffer, UINT64 count_buffer_offset);
+static Fn_ExecuteIndirect original_ExecuteIndirect = nullptr;
+static void STDMETHODCALLTYPE hook_ExecuteIndirect(ID3D12GraphicsCommandList* self, ID3D12CommandSignature *command_signature, UINT max_command_count, ID3D12Resource *arg_buffer, UINT64 arg_buffer_offset, ID3D12Resource *count_buffer, UINT64 count_buffer_offset);
+
+using Fn_AtomicCopyBufferUINT = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_buffer, UINT64 dst_offset, ID3D12Resource *src_buffer, UINT64 src_offset, UINT dependent_resource_count, ID3D12Resource *const *dependent_resources, const D3D12_SUBRESOURCE_RANGE_UINT64 *dependent_sub_resource_ranges);
+static Fn_AtomicCopyBufferUINT original_AtomicCopyBufferUINT = nullptr;
+static void STDMETHODCALLTYPE hook_AtomicCopyBufferUINT(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_buffer, UINT64 dst_offset, ID3D12Resource *src_buffer, UINT64 src_offset, UINT dependent_resource_count, ID3D12Resource *const *dependent_resources, const D3D12_SUBRESOURCE_RANGE_UINT64 *dependent_sub_resource_ranges) {
+    Access access(self);
+    return original_AtomicCopyBufferUINT(access.target(), dst_buffer, dst_offset, src_buffer, src_offset, dependent_resource_count, dependent_resources, dependent_sub_resource_ranges);
+}
+
+using Fn_AtomicCopyBufferUINT64 = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_buffer, UINT64 dst_offset, ID3D12Resource *src_buffer, UINT64 src_offset, UINT dependent_resource_count, ID3D12Resource *const *dependent_resources, const D3D12_SUBRESOURCE_RANGE_UINT64 *dependent_sub_resource_ranges);
+static Fn_AtomicCopyBufferUINT64 original_AtomicCopyBufferUINT64 = nullptr;
+static void STDMETHODCALLTYPE hook_AtomicCopyBufferUINT64(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_buffer, UINT64 dst_offset, ID3D12Resource *src_buffer, UINT64 src_offset, UINT dependent_resource_count, ID3D12Resource *const *dependent_resources, const D3D12_SUBRESOURCE_RANGE_UINT64 *dependent_sub_resource_ranges) {
+    Access access(self);
+    return original_AtomicCopyBufferUINT64(access.target(), dst_buffer, dst_offset, src_buffer, src_offset, dependent_resource_count, dependent_resources, dependent_sub_resource_ranges);
+}
+
+using Fn_OMSetDepthBounds = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, FLOAT min, FLOAT max);
+static Fn_OMSetDepthBounds original_OMSetDepthBounds = nullptr;
+static void STDMETHODCALLTYPE hook_OMSetDepthBounds(ID3D12GraphicsCommandList* self, FLOAT min, FLOAT max) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_OMSetDepthBounds(target, min, max);
+    });
+    return original_OMSetDepthBounds(access.target(), min, max);
+}
+
+using Fn_SetSamplePositions = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT sample_count, UINT pixel_count, D3D12_SAMPLE_POSITION *sample_positions);
+static Fn_SetSamplePositions original_SetSamplePositions = nullptr;
+static void STDMETHODCALLTYPE hook_SetSamplePositions(ID3D12GraphicsCommandList* self, UINT sample_count, UINT pixel_count, D3D12_SAMPLE_POSITION *sample_positions) {
+    Access access(self);
+    auto saved_sample_positions = copy_values(sample_positions, sample_count * pixel_count);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetSamplePositions(target, sample_count, pixel_count, values_pointer(saved_sample_positions));
+    });
+    return original_SetSamplePositions(access.target(), sample_count, pixel_count, sample_positions);
+}
+
+using Fn_ResolveSubresourceRegion = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_resource, UINT dst_sub_resource_idx, UINT dst_x, UINT dst_y, ID3D12Resource *src_resource, UINT src_sub_resource_idx, D3D12_RECT *src_rect, DXGI_FORMAT format, D3D12_RESOLVE_MODE mode);
+static Fn_ResolveSubresourceRegion original_ResolveSubresourceRegion = nullptr;
+static void STDMETHODCALLTYPE hook_ResolveSubresourceRegion(ID3D12GraphicsCommandList* self, ID3D12Resource *dst_resource, UINT dst_sub_resource_idx, UINT dst_x, UINT dst_y, ID3D12Resource *src_resource, UINT src_sub_resource_idx, D3D12_RECT *src_rect, DXGI_FORMAT format, D3D12_RESOLVE_MODE mode) {
+    Access access(self);
+    return original_ResolveSubresourceRegion(access.target(), dst_resource, dst_sub_resource_idx, dst_x, dst_y, src_resource, src_sub_resource_idx, src_rect, format, mode);
+}
+
+using Fn_SetViewInstanceMask = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT mask);
+static Fn_SetViewInstanceMask original_SetViewInstanceMask = nullptr;
+static void STDMETHODCALLTYPE hook_SetViewInstanceMask(ID3D12GraphicsCommandList* self, UINT mask) {
+    Access access(self);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetViewInstanceMask(target, mask);
+    });
+    return original_SetViewInstanceMask(access.target(), mask);
+}
+
+using Fn_WriteBufferImmediate = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT count, const D3D12_WRITEBUFFERIMMEDIATE_PARAMETER *parameters, const D3D12_WRITEBUFFERIMMEDIATE_MODE *modes);
+static Fn_WriteBufferImmediate original_WriteBufferImmediate = nullptr;
+static void STDMETHODCALLTYPE hook_WriteBufferImmediate(ID3D12GraphicsCommandList* self, UINT count, const D3D12_WRITEBUFFERIMMEDIATE_PARAMETER *parameters, const D3D12_WRITEBUFFERIMMEDIATE_MODE *modes) {
+    Access access(self);
+    return original_WriteBufferImmediate(access.target(), count, parameters, modes);
+}
+
+using Fn_SetProtectedResourceSession = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12ProtectedResourceSession *protected_resource_session);
+static Fn_SetProtectedResourceSession original_SetProtectedResourceSession = nullptr;
+static void STDMETHODCALLTYPE hook_SetProtectedResourceSession(ID3D12GraphicsCommandList* self, ID3D12ProtectedResourceSession *protected_resource_session) {
+    Access access(self);
+    auto saved_protected_resource_session = retain(protected_resource_session);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetProtectedResourceSession(target, saved_protected_resource_session.Get());
+    });
+    return original_SetProtectedResourceSession(access.target(), protected_resource_session);
+}
+
+using Fn_BeginRenderPass = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT render_targets_count, const D3D12_RENDER_PASS_RENDER_TARGET_DESC *render_targets, const D3D12_RENDER_PASS_DEPTH_STENCIL_DESC *depth_stencil, D3D12_RENDER_PASS_FLAGS flags);
+static Fn_BeginRenderPass original_BeginRenderPass = nullptr;
+static void STDMETHODCALLTYPE hook_BeginRenderPass(ID3D12GraphicsCommandList* self, UINT render_targets_count, const D3D12_RENDER_PASS_RENDER_TARGET_DESC *render_targets, const D3D12_RENDER_PASS_DEPTH_STENCIL_DESC *depth_stencil, D3D12_RENDER_PASS_FLAGS flags);
+
+using Fn_EndRenderPass = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self);
+static Fn_EndRenderPass original_EndRenderPass = nullptr;
+static void STDMETHODCALLTYPE hook_EndRenderPass(ID3D12GraphicsCommandList* self);
+
+using Fn_InitializeMetaCommand = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12MetaCommand *meta_command, const void *initialization_parameters_data, SIZE_T initialization_parameters_data_size_in_bytes);
+static Fn_InitializeMetaCommand original_InitializeMetaCommand = nullptr;
+static void STDMETHODCALLTYPE hook_InitializeMetaCommand(ID3D12GraphicsCommandList* self, ID3D12MetaCommand *meta_command, const void *initialization_parameters_data, SIZE_T initialization_parameters_data_size_in_bytes) {
+    Access access(self);
+    return original_InitializeMetaCommand(access.target(), meta_command, initialization_parameters_data, initialization_parameters_data_size_in_bytes);
+}
+
+using Fn_ExecuteMetaCommand = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12MetaCommand *meta_command, const void *execution_parameters_data, SIZE_T execution_parameters_data_size_in_bytes);
+static Fn_ExecuteMetaCommand original_ExecuteMetaCommand = nullptr;
+static void STDMETHODCALLTYPE hook_ExecuteMetaCommand(ID3D12GraphicsCommandList* self, ID3D12MetaCommand *meta_command, const void *execution_parameters_data, SIZE_T execution_parameters_data_size_in_bytes) {
+    Access access(self);
+    return original_ExecuteMetaCommand(access.target(), meta_command, execution_parameters_data, execution_parameters_data_size_in_bytes);
+}
+
+using Fn_BuildRaytracingAccelerationStructure = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, const D3D12_BUILD_RAYTRACING_ACCELERATION_STRUCTURE_DESC *desc, UINT postbuild_info_descs_count, const D3D12_RAYTRACING_ACCELERATION_STRUCTURE_POSTBUILD_INFO_DESC *postbuild_info_descs);
+static Fn_BuildRaytracingAccelerationStructure original_BuildRaytracingAccelerationStructure = nullptr;
+static void STDMETHODCALLTYPE hook_BuildRaytracingAccelerationStructure(ID3D12GraphicsCommandList* self, const D3D12_BUILD_RAYTRACING_ACCELERATION_STRUCTURE_DESC *desc, UINT postbuild_info_descs_count, const D3D12_RAYTRACING_ACCELERATION_STRUCTURE_POSTBUILD_INFO_DESC *postbuild_info_descs) {
+    Access access(self);
+    return original_BuildRaytracingAccelerationStructure(access.target(), desc, postbuild_info_descs_count, postbuild_info_descs);
+}
+
+using Fn_EmitRaytracingAccelerationStructurePostbuildInfo = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, const D3D12_RAYTRACING_ACCELERATION_STRUCTURE_POSTBUILD_INFO_DESC *desc, UINT src_acceleration_structures_count, const D3D12_GPU_VIRTUAL_ADDRESS *src_acceleration_structure_data);
+static Fn_EmitRaytracingAccelerationStructurePostbuildInfo original_EmitRaytracingAccelerationStructurePostbuildInfo = nullptr;
+static void STDMETHODCALLTYPE hook_EmitRaytracingAccelerationStructurePostbuildInfo(ID3D12GraphicsCommandList* self, const D3D12_RAYTRACING_ACCELERATION_STRUCTURE_POSTBUILD_INFO_DESC *desc, UINT src_acceleration_structures_count, const D3D12_GPU_VIRTUAL_ADDRESS *src_acceleration_structure_data) {
+    Access access(self);
+    return original_EmitRaytracingAccelerationStructurePostbuildInfo(access.target(), desc, src_acceleration_structures_count, src_acceleration_structure_data);
+}
+
+using Fn_CopyRaytracingAccelerationStructure = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, D3D12_GPU_VIRTUAL_ADDRESS dst_acceleration_structure_data, D3D12_GPU_VIRTUAL_ADDRESS src_acceleration_structure_data, D3D12_RAYTRACING_ACCELERATION_STRUCTURE_COPY_MODE mode);
+static Fn_CopyRaytracingAccelerationStructure original_CopyRaytracingAccelerationStructure = nullptr;
+static void STDMETHODCALLTYPE hook_CopyRaytracingAccelerationStructure(ID3D12GraphicsCommandList* self, D3D12_GPU_VIRTUAL_ADDRESS dst_acceleration_structure_data, D3D12_GPU_VIRTUAL_ADDRESS src_acceleration_structure_data, D3D12_RAYTRACING_ACCELERATION_STRUCTURE_COPY_MODE mode) {
+    Access access(self);
+    return original_CopyRaytracingAccelerationStructure(access.target(), dst_acceleration_structure_data, src_acceleration_structure_data, mode);
+}
+
+using Fn_SetPipelineState1 = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12StateObject *state_object);
+static Fn_SetPipelineState1 original_SetPipelineState1 = nullptr;
+static void STDMETHODCALLTYPE hook_SetPipelineState1(ID3D12GraphicsCommandList* self, ID3D12StateObject *state_object) {
+    Access access(self);
+    auto saved_state_object = retain(state_object);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_SetPipelineState1(target, saved_state_object.Get());
+    });
+    return original_SetPipelineState1(access.target(), state_object);
+}
+
+using Fn_DispatchRays = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, const D3D12_DISPATCH_RAYS_DESC *desc);
+static Fn_DispatchRays original_DispatchRays = nullptr;
+static void STDMETHODCALLTYPE hook_DispatchRays(ID3D12GraphicsCommandList* self, const D3D12_DISPATCH_RAYS_DESC *desc) {
+    Access access(self);
+    return original_DispatchRays(access.target(), desc);
+}
+
+using Fn_RSSetShadingRate = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, D3D12_SHADING_RATE base_shading_rate, const D3D12_SHADING_RATE_COMBINER *combiners);
+static Fn_RSSetShadingRate original_RSSetShadingRate = nullptr;
+static void STDMETHODCALLTYPE hook_RSSetShadingRate(ID3D12GraphicsCommandList* self, D3D12_SHADING_RATE base_shading_rate, const D3D12_SHADING_RATE_COMBINER *combiners) {
+    Access access(self);
+    auto saved_combiners = copy_values(combiners, 2);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_RSSetShadingRate(target, base_shading_rate, values_pointer(saved_combiners));
+    });
+    return original_RSSetShadingRate(access.target(), base_shading_rate, combiners);
+}
+
+using Fn_RSSetShadingRateImage = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, ID3D12Resource *shading_rate_image);
+static Fn_RSSetShadingRateImage original_RSSetShadingRateImage = nullptr;
+static void STDMETHODCALLTYPE hook_RSSetShadingRateImage(ID3D12GraphicsCommandList* self, ID3D12Resource *shading_rate_image) {
+    Access access(self);
+    auto saved_shading_rate_image = retain(shading_rate_image);
+    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {
+        original_RSSetShadingRateImage(target, saved_shading_rate_image.Get());
+    });
+    return original_RSSetShadingRateImage(access.target(), shading_rate_image);
+}
+
+using Fn_DispatchMesh = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT thread_group_count_x, UINT thread_group_count_y, UINT thread_group_count_z);
+static Fn_DispatchMesh original_DispatchMesh = nullptr;
+static void STDMETHODCALLTYPE hook_DispatchMesh(ID3D12GraphicsCommandList* self, UINT thread_group_count_x, UINT thread_group_count_y, UINT thread_group_count_z) {
+    Access access(self);
+    return original_DispatchMesh(access.target(), thread_group_count_x, thread_group_count_y, thread_group_count_z);
+}
+
+using Fn_Barrier = void(STDMETHODCALLTYPE*)(ID3D12GraphicsCommandList* self, UINT32 barrier_groups_count, const D3D12_BARRIER_GROUP *barrier_groups);
+static Fn_Barrier original_Barrier = nullptr;
+static void STDMETHODCALLTYPE hook_Barrier(ID3D12GraphicsCommandList* self, UINT32 barrier_groups_count, const D3D12_BARRIER_GROUP *barrier_groups);
+
+static void attach_recording_methods(void** table, unsigned version) {
+    if (version >= 0) attach_method(table[9], reinterpret_cast<void*>(hook_Close), reinterpret_cast<void**>(&original_Close), "Close");
+    if (version >= 0) attach_method(table[10], reinterpret_cast<void*>(hook_Reset), reinterpret_cast<void**>(&original_Reset), "Reset");
+    if (version >= 0) attach_method(table[11], reinterpret_cast<void*>(hook_ClearState), reinterpret_cast<void**>(&original_ClearState), "ClearState");
+    if (version >= 0) attach_method(table[12], reinterpret_cast<void*>(hook_DrawInstanced), reinterpret_cast<void**>(&original_DrawInstanced), "DrawInstanced");
+    if (version >= 0) attach_method(table[13], reinterpret_cast<void*>(hook_DrawIndexedInstanced), reinterpret_cast<void**>(&original_DrawIndexedInstanced), "DrawIndexedInstanced");
+    if (version >= 0) attach_method(table[14], reinterpret_cast<void*>(hook_Dispatch), reinterpret_cast<void**>(&original_Dispatch), "Dispatch");
+    if (version >= 0) attach_method(table[15], reinterpret_cast<void*>(hook_CopyBufferRegion), reinterpret_cast<void**>(&original_CopyBufferRegion), "CopyBufferRegion");
+    if (version >= 0) attach_method(table[16], reinterpret_cast<void*>(hook_CopyTextureRegion), reinterpret_cast<void**>(&original_CopyTextureRegion), "CopyTextureRegion");
+    if (version >= 0) attach_method(table[17], reinterpret_cast<void*>(hook_CopyResource), reinterpret_cast<void**>(&original_CopyResource), "CopyResource");
+    if (version >= 0) attach_method(table[18], reinterpret_cast<void*>(hook_CopyTiles), reinterpret_cast<void**>(&original_CopyTiles), "CopyTiles");
+    if (version >= 0) attach_method(table[19], reinterpret_cast<void*>(hook_ResolveSubresource), reinterpret_cast<void**>(&original_ResolveSubresource), "ResolveSubresource");
+    if (version >= 0) attach_method(table[20], reinterpret_cast<void*>(hook_IASetPrimitiveTopology), reinterpret_cast<void**>(&original_IASetPrimitiveTopology), "IASetPrimitiveTopology");
+    if (version >= 0) attach_method(table[21], reinterpret_cast<void*>(hook_RSSetViewports), reinterpret_cast<void**>(&original_RSSetViewports), "RSSetViewports");
+    if (version >= 0) attach_method(table[22], reinterpret_cast<void*>(hook_RSSetScissorRects), reinterpret_cast<void**>(&original_RSSetScissorRects), "RSSetScissorRects");
+    if (version >= 0) attach_method(table[23], reinterpret_cast<void*>(hook_OMSetBlendFactor), reinterpret_cast<void**>(&original_OMSetBlendFactor), "OMSetBlendFactor");
+    if (version >= 0) attach_method(table[24], reinterpret_cast<void*>(hook_OMSetStencilRef), reinterpret_cast<void**>(&original_OMSetStencilRef), "OMSetStencilRef");
+    if (version >= 0) attach_method(table[25], reinterpret_cast<void*>(hook_SetPipelineState), reinterpret_cast<void**>(&original_SetPipelineState), "SetPipelineState");
+    if (version >= 0) attach_method(table[26], reinterpret_cast<void*>(hook_ResourceBarrier), reinterpret_cast<void**>(&original_ResourceBarrier), "ResourceBarrier");
+    if (version >= 0) attach_method(table[27], reinterpret_cast<void*>(hook_ExecuteBundle), reinterpret_cast<void**>(&original_ExecuteBundle), "ExecuteBundle");
+    if (version >= 0) attach_method(table[28], reinterpret_cast<void*>(hook_SetDescriptorHeaps), reinterpret_cast<void**>(&original_SetDescriptorHeaps), "SetDescriptorHeaps");
+    if (version >= 0) attach_method(table[29], reinterpret_cast<void*>(hook_SetComputeRootSignature), reinterpret_cast<void**>(&original_SetComputeRootSignature), "SetComputeRootSignature");
+    if (version >= 0) attach_method(table[30], reinterpret_cast<void*>(hook_SetGraphicsRootSignature), reinterpret_cast<void**>(&original_SetGraphicsRootSignature), "SetGraphicsRootSignature");
+    if (version >= 0) attach_method(table[31], reinterpret_cast<void*>(hook_SetComputeRootDescriptorTable), reinterpret_cast<void**>(&original_SetComputeRootDescriptorTable), "SetComputeRootDescriptorTable");
+    if (version >= 0) attach_method(table[32], reinterpret_cast<void*>(hook_SetGraphicsRootDescriptorTable), reinterpret_cast<void**>(&original_SetGraphicsRootDescriptorTable), "SetGraphicsRootDescriptorTable");
+    if (version >= 0) attach_method(table[33], reinterpret_cast<void*>(hook_SetComputeRoot32BitConstant), reinterpret_cast<void**>(&original_SetComputeRoot32BitConstant), "SetComputeRoot32BitConstant");
+    if (version >= 0) attach_method(table[34], reinterpret_cast<void*>(hook_SetGraphicsRoot32BitConstant), reinterpret_cast<void**>(&original_SetGraphicsRoot32BitConstant), "SetGraphicsRoot32BitConstant");
+    if (version >= 0) attach_method(table[35], reinterpret_cast<void*>(hook_SetComputeRoot32BitConstants), reinterpret_cast<void**>(&original_SetComputeRoot32BitConstants), "SetComputeRoot32BitConstants");
+    if (version >= 0) attach_method(table[36], reinterpret_cast<void*>(hook_SetGraphicsRoot32BitConstants), reinterpret_cast<void**>(&original_SetGraphicsRoot32BitConstants), "SetGraphicsRoot32BitConstants");
+    if (version >= 0) attach_method(table[37], reinterpret_cast<void*>(hook_SetComputeRootConstantBufferView), reinterpret_cast<void**>(&original_SetComputeRootConstantBufferView), "SetComputeRootConstantBufferView");
+    if (version >= 0) attach_method(table[38], reinterpret_cast<void*>(hook_SetGraphicsRootConstantBufferView), reinterpret_cast<void**>(&original_SetGraphicsRootConstantBufferView), "SetGraphicsRootConstantBufferView");
+    if (version >= 0) attach_method(table[39], reinterpret_cast<void*>(hook_SetComputeRootShaderResourceView), reinterpret_cast<void**>(&original_SetComputeRootShaderResourceView), "SetComputeRootShaderResourceView");
+    if (version >= 0) attach_method(table[40], reinterpret_cast<void*>(hook_SetGraphicsRootShaderResourceView), reinterpret_cast<void**>(&original_SetGraphicsRootShaderResourceView), "SetGraphicsRootShaderResourceView");
+    if (version >= 0) attach_method(table[41], reinterpret_cast<void*>(hook_SetComputeRootUnorderedAccessView), reinterpret_cast<void**>(&original_SetComputeRootUnorderedAccessView), "SetComputeRootUnorderedAccessView");
+    if (version >= 0) attach_method(table[42], reinterpret_cast<void*>(hook_SetGraphicsRootUnorderedAccessView), reinterpret_cast<void**>(&original_SetGraphicsRootUnorderedAccessView), "SetGraphicsRootUnorderedAccessView");
+    if (version >= 0) attach_method(table[43], reinterpret_cast<void*>(hook_IASetIndexBuffer), reinterpret_cast<void**>(&original_IASetIndexBuffer), "IASetIndexBuffer");
+    if (version >= 0) attach_method(table[44], reinterpret_cast<void*>(hook_IASetVertexBuffers), reinterpret_cast<void**>(&original_IASetVertexBuffers), "IASetVertexBuffers");
+    if (version >= 0) attach_method(table[45], reinterpret_cast<void*>(hook_SOSetTargets), reinterpret_cast<void**>(&original_SOSetTargets), "SOSetTargets");
+    if (version >= 0) attach_method(table[46], reinterpret_cast<void*>(hook_OMSetRenderTargets), reinterpret_cast<void**>(&original_OMSetRenderTargets), "OMSetRenderTargets");
+    if (version >= 0) attach_method(table[47], reinterpret_cast<void*>(hook_ClearDepthStencilView), reinterpret_cast<void**>(&original_ClearDepthStencilView), "ClearDepthStencilView");
+    if (version >= 0) attach_method(table[48], reinterpret_cast<void*>(hook_ClearRenderTargetView), reinterpret_cast<void**>(&original_ClearRenderTargetView), "ClearRenderTargetView");
+    if (version >= 0) attach_method(table[49], reinterpret_cast<void*>(hook_ClearUnorderedAccessViewUint), reinterpret_cast<void**>(&original_ClearUnorderedAccessViewUint), "ClearUnorderedAccessViewUint");
+    if (version >= 0) attach_method(table[50], reinterpret_cast<void*>(hook_ClearUnorderedAccessViewFloat), reinterpret_cast<void**>(&original_ClearUnorderedAccessViewFloat), "ClearUnorderedAccessViewFloat");
+    if (version >= 0) attach_method(table[51], reinterpret_cast<void*>(hook_DiscardResource), reinterpret_cast<void**>(&original_DiscardResource), "DiscardResource");
+    if (version >= 0) attach_method(table[52], reinterpret_cast<void*>(hook_BeginQuery), reinterpret_cast<void**>(&original_BeginQuery), "BeginQuery");
+    if (version >= 0) attach_method(table[53], reinterpret_cast<void*>(hook_EndQuery), reinterpret_cast<void**>(&original_EndQuery), "EndQuery");
+    if (version >= 0) attach_method(table[54], reinterpret_cast<void*>(hook_ResolveQueryData), reinterpret_cast<void**>(&original_ResolveQueryData), "ResolveQueryData");
+    if (version >= 0) attach_method(table[55], reinterpret_cast<void*>(hook_SetPredication), reinterpret_cast<void**>(&original_SetPredication), "SetPredication");
+    if (version >= 0) attach_method(table[56], reinterpret_cast<void*>(hook_SetMarker), reinterpret_cast<void**>(&original_SetMarker), "SetMarker");
+    if (version >= 0) attach_method(table[57], reinterpret_cast<void*>(hook_BeginEvent), reinterpret_cast<void**>(&original_BeginEvent), "BeginEvent");
+    if (version >= 0) attach_method(table[58], reinterpret_cast<void*>(hook_EndEvent), reinterpret_cast<void**>(&original_EndEvent), "EndEvent");
+    if (version >= 0) attach_method(table[59], reinterpret_cast<void*>(hook_ExecuteIndirect), reinterpret_cast<void**>(&original_ExecuteIndirect), "ExecuteIndirect");
+    if (version >= 1) attach_method(table[60], reinterpret_cast<void*>(hook_AtomicCopyBufferUINT), reinterpret_cast<void**>(&original_AtomicCopyBufferUINT), "AtomicCopyBufferUINT");
+    if (version >= 1) attach_method(table[61], reinterpret_cast<void*>(hook_AtomicCopyBufferUINT64), reinterpret_cast<void**>(&original_AtomicCopyBufferUINT64), "AtomicCopyBufferUINT64");
+    if (version >= 1) attach_method(table[62], reinterpret_cast<void*>(hook_OMSetDepthBounds), reinterpret_cast<void**>(&original_OMSetDepthBounds), "OMSetDepthBounds");
+    if (version >= 1) attach_method(table[63], reinterpret_cast<void*>(hook_SetSamplePositions), reinterpret_cast<void**>(&original_SetSamplePositions), "SetSamplePositions");
+    if (version >= 1) attach_method(table[64], reinterpret_cast<void*>(hook_ResolveSubresourceRegion), reinterpret_cast<void**>(&original_ResolveSubresourceRegion), "ResolveSubresourceRegion");
+    if (version >= 1) attach_method(table[65], reinterpret_cast<void*>(hook_SetViewInstanceMask), reinterpret_cast<void**>(&original_SetViewInstanceMask), "SetViewInstanceMask");
+    if (version >= 2) attach_method(table[66], reinterpret_cast<void*>(hook_WriteBufferImmediate), reinterpret_cast<void**>(&original_WriteBufferImmediate), "WriteBufferImmediate");
+    if (version >= 3) attach_method(table[67], reinterpret_cast<void*>(hook_SetProtectedResourceSession), reinterpret_cast<void**>(&original_SetProtectedResourceSession), "SetProtectedResourceSession");
+    if (version >= 4) attach_method(table[68], reinterpret_cast<void*>(hook_BeginRenderPass), reinterpret_cast<void**>(&original_BeginRenderPass), "BeginRenderPass");
+    if (version >= 4) attach_method(table[69], reinterpret_cast<void*>(hook_EndRenderPass), reinterpret_cast<void**>(&original_EndRenderPass), "EndRenderPass");
+    if (version >= 4) attach_method(table[70], reinterpret_cast<void*>(hook_InitializeMetaCommand), reinterpret_cast<void**>(&original_InitializeMetaCommand), "InitializeMetaCommand");
+    if (version >= 4) attach_method(table[71], reinterpret_cast<void*>(hook_ExecuteMetaCommand), reinterpret_cast<void**>(&original_ExecuteMetaCommand), "ExecuteMetaCommand");
+    if (version >= 4) attach_method(table[72], reinterpret_cast<void*>(hook_BuildRaytracingAccelerationStructure), reinterpret_cast<void**>(&original_BuildRaytracingAccelerationStructure), "BuildRaytracingAccelerationStructure");
+    if (version >= 4) attach_method(table[73], reinterpret_cast<void*>(hook_EmitRaytracingAccelerationStructurePostbuildInfo), reinterpret_cast<void**>(&original_EmitRaytracingAccelerationStructurePostbuildInfo), "EmitRaytracingAccelerationStructurePostbuildInfo");
+    if (version >= 4) attach_method(table[74], reinterpret_cast<void*>(hook_CopyRaytracingAccelerationStructure), reinterpret_cast<void**>(&original_CopyRaytracingAccelerationStructure), "CopyRaytracingAccelerationStructure");
+    if (version >= 4) attach_method(table[75], reinterpret_cast<void*>(hook_SetPipelineState1), reinterpret_cast<void**>(&original_SetPipelineState1), "SetPipelineState1");
+    if (version >= 4) attach_method(table[76], reinterpret_cast<void*>(hook_DispatchRays), reinterpret_cast<void**>(&original_DispatchRays), "DispatchRays");
+    if (version >= 5) attach_method(table[77], reinterpret_cast<void*>(hook_RSSetShadingRate), reinterpret_cast<void**>(&original_RSSetShadingRate), "RSSetShadingRate");
+    if (version >= 5) attach_method(table[78], reinterpret_cast<void*>(hook_RSSetShadingRateImage), reinterpret_cast<void**>(&original_RSSetShadingRateImage), "RSSetShadingRateImage");
+    if (version >= 6) attach_method(table[79], reinterpret_cast<void*>(hook_DispatchMesh), reinterpret_cast<void**>(&original_DispatchMesh), "DispatchMesh");
+    if (version >= 7) attach_method(table[80], reinterpret_cast<void*>(hook_Barrier), reinterpret_cast<void**>(&original_Barrier), "Barrier");
+}

tools/windows/d3d12_command_hooks.h

@@ -0,0 +1,22 @@
+#pragma once
+#include "d3d12_external.h"
+#include <functional>
+#include <string>
+
+namespace d4r::win::commands {
+// Intercept only documented D3D12 COM methods on this device's runtime.
+// A boundary becomes prefix -> callback -> suffix at ExecuteCommandLists,
+// preserving preceding lists in the caller's batch and current-frame output.
+using Boundary = std::function<void(ID3D12CommandQueue*)>;
+void install(ID3D12Device* device);
+void record_boundary(ID3D12GraphicsCommandList* list, Boundary callback);
+std::string error();
+size_t live_recordings();
+D3D12_RESOURCE_STATES resource_state(ID3D12GraphicsCommandList* list, ID3D12Resource* resource, D3D12_RESOURCE_STATES fallback);
+// Internal copies and callback submissions must bypass logical list routing.
+struct InternalScope {
+    bool previous;
+    InternalScope();
+    ~InternalScope();
+};
+}

tools/windows/d3d12_command_probe.cpp

@@ -0,0 +1,163 @@
+#include "d3d12_external.h"
+#include <dxgi1_6.h>
+#include <d3dcompiler.h>
+#include <array>
+
+using namespace d4r::diag;
+using namespace d4r::win;
+namespace {
+constexpr unsigned count = 256, bytes = count * sizeof(uint32_t), seed = 0x12345;
+void drain(ID3D12Device* device, ID3D12CommandQueue* queue) {
+    ComPtr<ID3D12Fence> fence; dx(device->CreateFence(0, D3D12_FENCE_FLAG_NONE, IID_PPV_ARGS(fence.GetAddressOf())), "Probe fence");
+    Handle event; event.value = CreateEventW(nullptr, FALSE, FALSE, nullptr);
+    if (!event.value) throw std::runtime_error("Probe event");
+    dx(queue->Signal(fence.Get(), 1), "Probe signal");
+    if (fence->GetCompletedValue() < 1) {
+        dx(fence->SetEventOnCompletion(1, event.value), "Probe fence event");
+        if (WaitForSingleObject(event.value, 30000) != WAIT_OBJECT_0) throw std::runtime_error("Probe GPU timeout");
+    }
+}
+struct NativeList {
+    ComPtr<ID3D12CommandAllocator> allocator;
+    ComPtr<ID3D12GraphicsCommandList> list;
+    explicit NativeList(ID3D12Device* device) {
+        dx(device->CreateCommandAllocator(D3D12_COMMAND_LIST_TYPE_DIRECT, IID_PPV_ARGS(allocator.GetAddressOf())), "Probe allocator");
+        dx(device->CreateCommandList(0, D3D12_COMMAND_LIST_TYPE_DIRECT, allocator.Get(), nullptr, IID_PPV_ARGS(list.GetAddressOf())), "Probe command list");
+    }
+    void submit(ID3D12Device* device, ID3D12CommandQueue* queue) {
+        dx(list->Close(), "Close probe list"); ID3D12CommandList* lists[] = {list.Get()};
+        queue->ExecuteCommandLists(1, lists); drain(device, queue);
+    }
+};
+ComPtr<ID3D12Resource> output_buffer(ID3D12Device* device) {
+    D3D12_HEAP_PROPERTIES heap{}; heap.Type = D3D12_HEAP_TYPE_DEFAULT; heap.CreationNodeMask = heap.VisibleNodeMask = 1;
+    D3D12_RESOURCE_DESC desc{}; desc.Dimension = D3D12_RESOURCE_DIMENSION_BUFFER; desc.Width = bytes;
+    desc.Height = desc.DepthOrArraySize = desc.MipLevels = desc.SampleDesc.Count = 1;
+    desc.Layout = D3D12_TEXTURE_LAYOUT_ROW_MAJOR; desc.Flags = D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS;
+    ComPtr<ID3D12Resource> result;
+    dx(device->CreateCommittedResource(&heap, D3D12_HEAP_FLAG_NONE, &desc, D3D12_RESOURCE_STATE_UNORDERED_ACCESS,
+        nullptr, IID_PPV_ARGS(result.GetAddressOf())), "Probe UAV");
+    return result;
+}
+void upload(ID3D12Resource* buffer, uint32_t multiplier, uint32_t addend) {
+    void* data = nullptr; D3D12_RANGE noRead{}; dx(buffer->Map(0, &noRead, &data), "Probe upload map");
+    auto* values = static_cast<uint32_t*>(data);
+    for (unsigned i = 0; i < count; ++i) values[i] = i * multiplier + addend;
+    buffer->Unmap(0, nullptr);
+}
+bool verify(ID3D12Resource* buffer, uint32_t multiplier, uint32_t addend) {
+    void* data = nullptr; D3D12_RANGE range{0, bytes}; dx(buffer->Map(0, &range, &data), "Probe verification map");
+    const auto* values = static_cast<const uint32_t*>(data);
+    bool passed = true;
+    for (unsigned i = 0; i < count; ++i) if (values[i] != i * multiplier + addend + seed) {
+        std::fprintf(stderr, "COMMAND_DIFFERENT index=%u expected=%u actual=%u\n", i, i * multiplier + addend + seed, values[i]);
+        passed = false; break;
+    }
+    D3D12_RANGE noWrite{}; buffer->Unmap(0, &noWrite); return passed;
+}
+struct Callback {
+    ID3D12Device* device;
+    ID3D12Resource* input;
+    ID3D12Resource* output;
+    ID3D12Resource* replacement;
+    ID3D12Resource* readback;
+    bool prefix_verified = false;
+};
+void WINAPI boundary(ID3D12CommandQueue* queue, void* context) {
+    auto& callback = *static_cast<Callback*>(context);
+    NativeList list(callback.device);
+    transition(list.list.Get(), callback.output, D3D12_RESOURCE_STATE_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_COPY_SOURCE);
+    list.list->CopyResource(callback.readback, callback.output);
+    transition(list.list.Get(), callback.output, D3D12_RESOURCE_STATE_COPY_SOURCE, D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
+    transition(list.list.Get(), callback.input, D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE, D3D12_RESOURCE_STATE_COPY_DEST);
+    list.list->CopyResource(callback.input, callback.replacement);
+    transition(list.list.Get(), callback.input, D3D12_RESOURCE_STATE_COPY_DEST, D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE);
+    list.submit(callback.device, queue);
+    callback.prefix_verified = verify(callback.readback, 17, 10);
+}
+}
+
+int main(int argc, char** argv) {
+    start();
+    try {
+        Args args(argc, argv);
+        if (args.module.empty()) throw std::runtime_error("--module d4r_nvngx.dll is required; NVIDIA DLLs are not needed");
+        HipApi hip(args.hip_root); hipDeviceProp_t props{}; hip.select_gfx1201(args.device, props);
+        ComPtr<IDXGIFactory4> factory; dx(CreateDXGIFactory1(IID_PPV_ARGS(factory.GetAddressOf())), "Probe DXGI");
+        ComPtr<IDXGIAdapter1> adapter;
+        for (UINT i = 0; ; ++i) {
+            dx(factory->EnumAdapters1(i, adapter.ReleaseAndGetAddressOf()), "Probe adapter");
+            DXGI_ADAPTER_DESC1 desc{}; dx(adapter->GetDesc1(&desc), "Probe adapter desc");
+            if (!std::memcmp(&desc.AdapterLuid, props.luid, sizeof(LUID))) break;
+        }
+        ComPtr<ID3D12Device> device;
+        dx(D3D12CreateDevice(adapter.Get(), D3D_FEATURE_LEVEL_12_0, IID_PPV_ARGS(device.GetAddressOf())), "Probe D3D12");
+        ComPtr<ID3D12CommandQueue> queue;
+        D3D12_COMMAND_QUEUE_DESC qd{}; qd.Type = D3D12_COMMAND_LIST_TYPE_DIRECT;
+        dx(device->CreateCommandQueue(&qd, IID_PPV_ARGS(queue.GetAddressOf())), "Probe queue");
+        wchar_t system[32768]{}; if (!GetSystemDirectoryW(system, 32768)) throw std::runtime_error("System directory");
+        Library compiler(std::filesystem::path(system) / L"d3dcompiler_47.dll");
+        auto compile = compiler.symbol<decltype(&D3DCompile)>("D3DCompile");
+        static constexpr char shader[] = "RWStructuredBuffer<uint> Output : register(u0); StructuredBuffer<uint> Input : register(t0); cbuffer Constants : register(b0) { uint Seed; }; [numthreads(64,1,1)] void main(uint3 t:SV_DispatchThreadID) { Output[t.x] = Input[t.x] + Seed; }";
+        ComPtr<ID3DBlob> code, errors;
+        dx(compile(shader, sizeof(shader)-1, nullptr, nullptr, nullptr, "main", "cs_5_1", D3DCOMPILE_OPTIMIZATION_LEVEL3,
+            0, code.GetAddressOf(), errors.GetAddressOf()), "Probe HLSL");
+        D3D12_ROOT_PARAMETER roots[3]{};
+        roots[0].ParameterType = D3D12_ROOT_PARAMETER_TYPE_UAV; roots[0].Descriptor.ShaderRegister = 0;
+        roots[1].ParameterType = D3D12_ROOT_PARAMETER_TYPE_SRV; roots[1].Descriptor.ShaderRegister = 0;
+        roots[2].ParameterType = D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS; roots[2].Constants.Num32BitValues = 1;
+        D3D12_ROOT_SIGNATURE_DESC signature{}; signature.NumParameters = 3; signature.pParameters = roots;
+        ComPtr<ID3DBlob> serialized;
+        dx(D3D12SerializeRootSignature(&signature, D3D_ROOT_SIGNATURE_VERSION_1, serialized.GetAddressOf(), errors.ReleaseAndGetAddressOf()), "Probe root signature serialize");
+        ComPtr<ID3D12RootSignature> root;
+        dx(device->CreateRootSignature(0, serialized->GetBufferPointer(), serialized->GetBufferSize(), IID_PPV_ARGS(root.GetAddressOf())), "Probe root signature");
+        D3D12_COMPUTE_PIPELINE_STATE_DESC pipeline{};
+        pipeline.pRootSignature = root.Get(); pipeline.CS = {code->GetBufferPointer(), code->GetBufferSize()};
+        ComPtr<ID3D12PipelineState> pso; dx(device->CreateComputePipelineState(&pipeline, IID_PPV_ARGS(pso.GetAddressOf())), "Probe PSO");
+        Library shim(wide(args.module));
+        using Install = unsigned(*)(ID3D12Device*);
+        using Record = unsigned(*)(ID3D12GraphicsCommandList*, void(WINAPI*)(ID3D12CommandQueue*, void*), void*);
+        using Live = unsigned long long(*)();
+        if (shim.symbol<Install>("d4r_D3D12_InstallCommandBackend")(device.Get()) != 1) throw std::runtime_error("Command backend installation failed");
+        auto record = shim.symbol<Record>("d4r_D3D12_RecordDiagnosticBoundary");
+        auto live = shim.symbol<Live>("d4r_D3D12_LiveRecordings");
+        for (unsigned iteration = 0; iteration < args.iterations; ++iteration) {
+            auto input = make_buffer(device.Get(), bytes, D3D12_HEAP_TYPE_DEFAULT, D3D12_HEAP_FLAG_NONE, D3D12_RESOURCE_STATE_COPY_DEST);
+            auto output = output_buffer(device.Get());
+            auto first = make_buffer(device.Get(), bytes, D3D12_HEAP_TYPE_UPLOAD, D3D12_HEAP_FLAG_NONE, D3D12_RESOURCE_STATE_GENERIC_READ);
+            auto second = make_buffer(device.Get(), bytes, D3D12_HEAP_TYPE_UPLOAD, D3D12_HEAP_FLAG_NONE, D3D12_RESOURCE_STATE_GENERIC_READ);
+            auto readback = make_buffer(device.Get(), bytes, D3D12_HEAP_TYPE_READBACK, D3D12_HEAP_FLAG_NONE, D3D12_RESOURCE_STATE_COPY_DEST);
+            auto prefix = make_buffer(device.Get(), bytes, D3D12_HEAP_TYPE_READBACK, D3D12_HEAP_FLAG_NONE, D3D12_RESOURCE_STATE_COPY_DEST);
+            upload(first.Get(), 17, 10); upload(second.Get(), 25, 100);
+            {
+                NativeList a(device.Get()), b(device.Get()), c(device.Get());
+                a.list->CopyResource(input.Get(), first.Get());
+                transition(a.list.Get(), input.Get(), D3D12_RESOURCE_STATE_COPY_DEST, D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE);
+                dx(a.list->Close(), "Close predecessor A");
+                b.list->SetComputeRootSignature(root.Get()); b.list->SetPipelineState(pso.Get());
+                b.list->SetComputeRootUnorderedAccessView(0, output->GetGPUVirtualAddress());
+                b.list->SetComputeRootShaderResourceView(1, input->GetGPUVirtualAddress());
+                uint32_t temporary = seed;
+                b.list->SetComputeRoot32BitConstants(2, 1, &temporary, 0);
+                temporary = 0xdeadbeef; // Deep-copy check: replay must retain Seed.
+                b.list->Dispatch(count / 64, 1, 1);
+                Callback callback{device.Get(), input.Get(), output.Get(), second.Get(), prefix.Get()};
+                if (record(b.list.Get(), boundary, &callback) != 1) throw std::runtime_error("Record split failed");
+                b.list->Dispatch(count / 64, 1, 1); // No rebinding after the split.
+                dx(b.list->Close(), "Close logical list B");
+                transition(c.list.Get(), output.Get(), D3D12_RESOURCE_STATE_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_COPY_SOURCE);
+                c.list->CopyResource(readback.Get(), output.Get());
+                transition(c.list.Get(), output.Get(), D3D12_RESOURCE_STATE_COPY_SOURCE, D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
+                dx(c.list->Close(), "Close consumer C");
+                ID3D12CommandList* batch[] = {a.list.Get(), b.list.Get(), c.list.Get()};
+                queue->ExecuteCommandLists(3, batch); drain(device.Get(), queue.Get());
+                if (!callback.prefix_verified || !verify(readback.Get(), 25, 100)) throw std::runtime_error("Queue order or suffix root state mismatch");
+            }
+            if (live() != 0) throw std::runtime_error("D3D12 recording metadata leaked after command-list Release");
+        }
+        std::printf("PASS D3D12_COMMAND_BACKEND architecture=gfx1201 iterations=%u batch_order=1 root_state=1 deep_copy=1 live_recordings=0 frame_age=0\n", args.iterations);
+        return 0;
+    } catch (const std::exception& failure) {
+        std::fprintf(stderr, "FAIL D3D12_COMMAND_BACKEND %s\n", failure.what()); loaded_modules(); return 4;
+    }
+}

tools/windows/d3d12_dlss_probe.cpp

@@ -16,6 +16,8 @@ int main(int argc, char** argv) {
         if (!diagDir && args.interop_mode != "images") throw std::runtime_error("Set D4R_DIAG_DIR");
         const std::filesystem::path directory(diagDir ? wide(diagDir) : std::filesystem::current_path().wstring());
         std::filesystem::create_directories(directory);
+        const bool commandBackend = args.interop_mode == "command-list";
+        if (commandBackend && _putenv_s("D4R_D3D12_COMMAND_BACKEND", "1")) throw std::runtime_error("Cannot enable command backend");
         for (const auto& pair : {std::make_pair("D4R_HIP_ROOT", args.hip_root),
              std::make_pair("D4R_NVCUDA_DLL", args.cuda_dll), std::make_pair("D4R_NVAPI_DLL", args.nvapi_dll),
              std::make_pair("D4R_NGX_CORE", args.ngx_core), std::make_pair("D4R_DLSS_DLL", args.dlss_dll)})
@@ -93,7 +95,7 @@ int main(int argc, char** argv) {
                 i == 3 ? D3D12_RESOURCE_STATE_UNORDERED_ACCESS : D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE);
         }
         dx(list->Close(), "Close(synthetic producers)"); ID3D12CommandList* producer[] = {list.Get()};
-        queue->ExecuteCommandLists(1, producer); drain();
+        if (!commandBackend) { queue->ExecuteCommandLists(1, producer); drain(); }
         if (args.interop_mode == "images") {
             CudaApi cuda(args.cuda_dll); cuda.check(cuda.cuInit(0), "cuInit(image interop)");
             CUdevice ordinal = 0; CUcontext context = nullptr;
@@ -164,17 +166,43 @@ int main(int argc, char** argv) {
         for (unsigned i = 0; i < 5; ++i) d4r_ngx_set_d3d12_resource(parameters, names[i], textures[i].image.Get());
         d4r_ngx_set_int(parameters, "Disable.Watermark", 1);
         auto readback = make_buffer(device.Get(), textures[3].bytes, D3D12_HEAP_TYPE_READBACK, D3D12_HEAP_FLAG_NONE, D3D12_RESOURCE_STATE_COPY_DEST);
+        ComPtr<ID3D12GraphicsCommandList> predecessor;
+        ComPtr<ID3D12CommandAllocator> predecessorAllocator;
+        if (commandBackend) {
+            // Keep initial synthetic producers unsubmitted until the first
+            // batch, before the list which contains the NGX evaluation.
+            predecessor = list; predecessorAllocator = allocator;
+            dx(device->CreateCommandAllocator(D3D12_COMMAND_LIST_TYPE_DIRECT, IID_PPV_ARGS(allocator.ReleaseAndGetAddressOf())), "Create game-style allocator");
+            dx(device->CreateCommandList(0, D3D12_COMMAND_LIST_TYPE_DIRECT, allocator.Get(), nullptr,
+                IID_PPV_ARGS(list.ReleaseAndGetAddressOf())), "Create game-style command list");
+            dx(list->Close(), "Close initial game-style list");
+        }
         for (unsigned frame = 0; frame < args.iterations; ++frame) {
             d4r_ngx_set_int(parameters, "Reset", frame == 0 ? 1 : 0);
-            check(shim.symbol<Evaluate>("d4r_D3D12_EvaluateAtBoundary")(queue.Get(), handle, parameters), "D3D12 same-frame Evaluate");
+            if (commandBackend) {
-            dx(allocator->Reset(), "Reset(harness allocator)"); dx(list->Reset(allocator.Get(), nullptr), "Reset(harness list)");
+                dx(allocator->Reset(), "Reset(game-style allocator)"); dx(list->Reset(allocator.Get(), nullptr), "Reset(game-style list)");
+                using RecordedEvaluate = unsigned(*)(ID3D12GraphicsCommandList*, void*, void*, void*);
+                check(shim.symbol<RecordedEvaluate>("NVSDK_NGX_D3D12_EvaluateFeature")(list.Get(), handle, parameters, nullptr), "D3D12 recorded same-frame Evaluate");
+                // The caller may reuse/mutate its parameter object as soon as
+                // Evaluate returns. The queued snapshot must remain immutable.
+                d4r_ngx_set_d3d12_resource(parameters, "Color", nullptr);
+                d4r_ngx_set_int(parameters, "Reset", 0);
+            } else {
+                check(shim.symbol<Evaluate>("d4r_D3D12_EvaluateAtBoundary")(queue.Get(), handle, parameters), "D3D12 same-frame Evaluate");
+                dx(allocator->Reset(), "Reset(harness allocator)"); dx(list->Reset(allocator.Get(), nullptr), "Reset(harness list)");
+            }
             transition(list.Get(), textures[3].image.Get(), D3D12_RESOURCE_STATE_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_COPY_SOURCE);
             D3D12_TEXTURE_COPY_LOCATION src{}, dst{};
             src.pResource = textures[3].image.Get(); src.Type = D3D12_TEXTURE_COPY_TYPE_SUBRESOURCE_INDEX;
             dst.pResource = readback.Get(); dst.Type = D3D12_TEXTURE_COPY_TYPE_PLACED_FOOTPRINT; dst.PlacedFootprint = textures[3].footprint;
             list->CopyTextureRegion(&dst, 0, 0, 0, &src, nullptr);
             transition(list.Get(), textures[3].image.Get(), D3D12_RESOURCE_STATE_COPY_SOURCE, D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
-            dx(list->Close(), "Close(verification)"); ID3D12CommandList* verify[] = {list.Get()}; queue->ExecuteCommandLists(1, verify); drain();
+            dx(list->Close(), "Close(verification)");
+            if (commandBackend && frame == 0) {
+                ID3D12CommandList* batch[] = {predecessor.Get(), list.Get()}; queue->ExecuteCommandLists(2, batch);
+            } else { ID3D12CommandList* verify[] = {list.Get()}; queue->ExecuteCommandLists(1, verify); }
+            drain();
+            if (commandBackend) d4r_ngx_set_d3d12_resource(parameters, "Color", textures[0].image.Get());
             void* raw = nullptr; D3D12_RANGE range{0, size_t(textures[3].bytes)};
             dx(readback->Map(0, &range, &raw), "Map(verification only)");
             std::vector<uint16_t> pixels(size_t(outWidth) * outHeight * 4);
@@ -193,7 +221,8 @@ int main(int argc, char** argv) {
             if (variance < 1e-6) throw std::runtime_error("D3D12 output lost the synthetic pattern");
         }
         loaded_modules();
-        std::printf("PASS D3D12_DLSS architecture=gfx1201 preset=%u frames=%u fast_path_cpu_copies=0 frame_age=0 queue_integration=explicit_harness\n", args.preset, args.iterations);
+        std::printf("PASS D3D12_DLSS architecture=gfx1201 preset=%u frames=%u fast_path_cpu_copies=0 frame_age=0 queue_integration=%s\n",
+            args.preset, args.iterations, commandBackend ? "recorded_command_list" : "explicit_harness");
         return 0;
     } catch (const std::exception& error) { std::fprintf(stderr, "FAIL D3D12_DLSS %s\n", error.what()); loaded_modules(); return 4; }
 }

tools/windows/d4r_nvngx_windows.cpp

@@ -44,6 +44,8 @@ API unsigned NVSDK_NGX_D3D12_Init_Ext(unsigned long long app, const wchar_t* dat
     std::lock_guard<std::mutex> lock(apiMutex);
     return call([&] {
         if (!runtime) runtime = std::make_shared<d4r::win::Runtime>(device, app, data, sdk);
+        if (const char* backend = std::getenv("D4R_D3D12_COMMAND_BACKEND"); backend && std::string(backend) == "1")
+            d4r::win::commands::install(device);
         return 1u;
     });
 }
@@ -84,10 +86,33 @@ API unsigned d4r_D3D12_EvaluateAtBoundary(ID3D12CommandQueue* queue, Handle* han
       auto it = features.find(handle); if (it == features.end() || !p) return invalid; feature = it->second; }
     return call([&] { feature->evaluate_boundary(queue, p); return 1u; });
 }
-API unsigned NVSDK_NGX_D3D12_EvaluateFeature(ID3D12GraphicsCommandList*, const Handle*, void*, void*) {
+API unsigned NVSDK_NGX_D3D12_EvaluateFeature(ID3D12GraphicsCommandList* list, const Handle* handle, void* p, void*) {
-    // Explicitly refuse an unsubmitted game list until queue integration exists.
+    std::shared_ptr<d4r::win::Feature> feature;
-    std::fprintf(stderr, "D4R_WINDOWS_FAILURE command-list queue backend pending; no frame-age fallback\n");
+    { std::lock_guard<std::mutex> lock(apiMutex);
-    return failure;
+      auto it = features.find(const_cast<Handle*>(handle));
+      if (it == features.end() || !p || !list) return invalid; feature = it->second; }
+    return call([&] {
+        struct Snapshot {
+            void* parameters = d4r_ngx_parameters_create();
+            std::vector<d4r::win::ComPtr<ID3D12Resource>> resources;
+            ~Snapshot() { if (parameters) d4r_ngx_parameters_destroy(parameters); }
+        };
+        auto snapshot = std::make_shared<Snapshot>();
+        if (!snapshot->parameters) return failure;
+        d4r::ngx::copy_create(p, snapshot->parameters); d4r::ngx::copy_frame(p, snapshot->parameters);
+        const char* names[] = {"Color", "Depth", "MotionVectors", "Output", "ExposureTexture"};
+        const char* states[] = {"D4R.Color.State", "D4R.Depth.State", "D4R.Motion.State", "D4R.Output.State", "D4R.Exposure.State"};
+        for (unsigned i = 0; i < 5; ++i) {
+            ID3D12Resource* resource = nullptr; (void)d4r_ngx_get_d3d12_resource(p, names[i], &resource);
+            snapshot->resources.emplace_back(resource);
+            d4r_ngx_set_d3d12_resource(snapshot->parameters, names[i], resource);
+            const auto state = d4r::win::commands::resource_state(list, resource, D3D12_RESOURCE_STATES(
+                d4r::ngx::uint_value(p, states[i], i == 3 ? D3D12_RESOURCE_STATE_UNORDERED_ACCESS : D3D12_RESOURCE_STATE_NON_PIXEL_SHADER_RESOURCE)));
+            d4r_ngx_set_uint(snapshot->parameters, states[i], unsigned(state));
+        }
+        d4r::win::commands::record_boundary(list, [feature, snapshot](ID3D12CommandQueue* queue) { feature->evaluate_boundary(queue, snapshot->parameters); });
+        return 1u;
+    });
 }
 API unsigned NVSDK_NGX_D3D12_EvaluateFeature_C(ID3D12GraphicsCommandList* list, const Handle* handle, void* p, void* callback) {
     return NVSDK_NGX_D3D12_EvaluateFeature(list, handle, p, callback);
@@ -109,3 +134,14 @@ API unsigned NVSDK_NGX_D3D12_Shutdown() {
     });
 }
 API unsigned NVSDK_NGX_D3D12_Shutdown1(ID3D12Device*) { return NVSDK_NGX_D3D12_Shutdown(); }
+
+// Public diagnostics exercise queue/list integration without NVIDIA binaries.
+API unsigned d4r_D3D12_InstallCommandBackend(ID3D12Device* device) {
+    return call([&] { d4r::win::commands::install(device); return 1u; });
+}
+using DiagnosticBoundary = void(WINAPI*)(ID3D12CommandQueue*, void*);
+API unsigned d4r_D3D12_RecordDiagnosticBoundary(ID3D12GraphicsCommandList* list, DiagnosticBoundary callback, void* context) {
+    if (!callback) return invalid;
+    return call([&] { d4r::win::commands::record_boundary(list, [=](ID3D12CommandQueue* queue) { callback(queue, context); }); return 1u; });
+}
+API unsigned long long d4r_D3D12_LiveRecordings() { return d4r::win::commands::live_recordings(); }

tools/windows/generate_d3d12_hooks.py

@@ -0,0 +1,101 @@
+"""Generate typed command forwarding from public MinGW D3D12 SDK declarations.
+
+Run with --header <SDK/include/d3d12.h> --output d3d12_command_hooks.generated.h.
+Every recording method through CommandList7 is forwarded. State setters copy
+their host arguments; COM object arguments retain a reference. SDK slot numbers
+are derived from the public interface inheritance, never a driver binary.
+"""
+import argparse
+import pathlib
+import re
+
+POINTER_COUNTS = {
+    'RSSetViewports': {'viewports': 'viewport_count'},
+    'RSSetScissorRects': {'rects': 'rect_count'},
+    'OMSetBlendFactor': {'blend_factor': '4'},
+    'SetDescriptorHeaps': {'heaps': 'heap_count'},
+    'SetComputeRoot32BitConstants': {'data': 'constant_count'},
+    'SetGraphicsRoot32BitConstants': {'data': 'constant_count'},
+    'IASetIndexBuffer': {'view': '1'},
+    'IASetVertexBuffers': {'views': 'view_count'},
+    'SOSetTargets': {'views': 'view_count'},
+    'OMSetRenderTargets': {'render_target_descriptors': '(single_descriptor_handle ? 1 : render_target_descriptor_count)', 'depth_stencil_descriptor': '1'},
+    'SetSamplePositions': {'sample_positions': 'sample_count * pixel_count'},
+    'RSSetShadingRate': {'combiners': '2'},
+}
+SPECIAL = {'Close', 'Reset', 'ClearState', 'ResourceBarrier', 'Barrier',
+           'BeginEvent', 'EndEvent', 'BeginRenderPass', 'EndRenderPass',
+           'BeginQuery', 'EndQuery', 'ExecuteBundle', 'ExecuteIndirect', 'SetPredication'}
+
+
+def generate(source):
+    classes = []
+    for match in re.finditer(r'\n(ID3D12GraphicsCommandList\d*) : public \w+\s*\{(.*?)\n\};', source, re.S):
+        name, body = match.groups()
+        methods = re.findall(r'virtual\s+(.*?)\s+STDMETHODCALLTYPE\s+(\w+)\s*\((.*?)\)\s*= 0;', body, re.S)
+        classes.append((int(name[len('ID3D12GraphicsCommandList'):] or '0'), name, methods))
+    if [version for version, _, _ in sorted(classes)] != list(range(8)):
+        raise RuntimeError('Expected exactly the public CommandList0..7 interfaces; review a changed SDK')
+    lines = ['// Generated by generate_d3d12_hooks.py from the public SDK. Do not edit.', '']
+    registrations = []
+    slot = 9  # IUnknown(3), ID3D12Object(4), DeviceChild(1), CommandList(1).
+    for version, interface, methods in sorted(classes):
+        for result, name, arguments in methods:
+            result = ' '.join(result.split())
+            arguments = ' '.join(arguments.split())
+            declarations = [arg.strip() for arg in arguments.split(',')] if arguments else []
+            names = []
+            for declaration in declarations:
+                match = re.search(r'(\w+)\s*(?:\[\d+\])?$', declaration)
+                if not match:
+                    raise RuntimeError(f'Unparsed SDK argument: {name}: {declaration}')
+                names.append(match[1])
+            signature = ', '.join(['ID3D12GraphicsCommandList* self'] + declarations)
+            call = ', '.join(['access.target()'] + names)
+            lines.append(f'using Fn_{name} = {result}(STDMETHODCALLTYPE*)({signature});')
+            lines.append(f'static Fn_{name} original_{name} = nullptr;')
+            if name in SPECIAL:
+                lines.append(f'static {result} STDMETHODCALLTYPE hook_{name}({signature});')
+            else:
+                lines += [f'static {result} STDMETHODCALLTYPE hook_{name}({signature}) {{', '    Access access(self);']
+                stateful = name.startswith(('Set', 'IASet', 'RSSet', 'OMSet', 'SOSet')) and name != 'SetMarker'
+                if stateful:
+                    captured = []
+                    for declaration, argument in zip(declarations, names):
+                        pointer = '*' in declaration or '[' in declaration
+                        if pointer and 'ID3D12' in declaration:
+                            if declaration.count('*') > 1:
+                                count = POINTER_COUNTS.get(name, {}).get(argument)
+                                if count is None:
+                                    raise RuntimeError(f'No COM array copy rule: {name}: {declaration}')
+                                lines.append(f'    auto saved_{argument} = retain_values({argument}, {count});')
+                                captured.append(f'saved_{argument}.raw.data()')
+                            else:
+                                lines.append(f'    auto saved_{argument} = retain({argument});')
+                                captured.append(f'saved_{argument}.Get()')
+                        elif pointer:
+                            count = POINTER_COUNTS.get(name, {}).get(argument)
+                            if count is None:
+                                raise RuntimeError(f'No deep-copy rule for {name}: {declaration}')
+                            value = f'static_cast<const UINT*>({argument})' if argument == 'data' else argument
+                            lines.append(f'    auto saved_{argument} = copy_values({value}, {count});')
+                            captured.append(f'values_pointer(saved_{argument})')
+                        else:
+                            captured.append(argument)
+                    lines += ['    if (access.recording) access.recording->state.emplace_back([=](ID3D12GraphicsCommandList* target) mutable {',
+                              f'        original_{name}(' + ', '.join(['target'] + captured) + ');', '    });']
+                lines += [f'    return original_{name}({call});', '}']
+            registrations.append(f'    attach_method(table[{slot}], reinterpret_cast<void*>(hook_{name}), reinterpret_cast<void**>(&original_{name}), "{name}");')
+            registrations[-1] = f'    if (version >= {version})' + registrations[-1][3:]
+            slot += 1
+            lines.append('')
+    lines += ['static void attach_recording_methods(void** table, unsigned version) {'] + registrations + ['}', '']
+    return '\n'.join(lines)
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('--header', type=pathlib.Path, required=True)
+    parser.add_argument('--output', type=pathlib.Path, required=True)
+    args = parser.parse_args()
+    args.output.write_text(generate(args.header.read_text()), encoding='utf-8')

tools/windows/ngx_windows_runtime.h

@@ -2,6 +2,7 @@
 #include "d3d12_external.h"
 #include "cuda_image_api.h"
 #include "ngx_parameters.h"
+#include "d3d12_command_hooks.h"
 #include <chrono>
 #include <memory>
 #include <mutex>
@@ -160,6 +161,7 @@ public:
     // Explicit harness boundary: all earlier producers must already be queued.
     // The forthcoming command-list backend calls the same stages at queue submit.
     void evaluate_boundary(ID3D12CommandQueue* queue, void* parameters) {
+        commands::InternalScope internal;
         std::lock_guard<std::mutex> lock(rt_->mutex);
         if (!queue) throw std::runtime_error("Null D3D12 queue");
         if (queue_ && queue_.Get() != queue) throw std::runtime_error("A feature cannot switch queues without an explicit drain");