[video_core] cpu buffer fix + kepler uploads / maxwell macro dirty tracking fixes (#4473)

- [x] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions).
- [x] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md)
- [x] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability.

-------------------

Aimed to fix two known UE5 crashes: Kepler uploads and Maxwell macros, both caused by CPU/GPU races due dirty tracking issues.

Kepler ComputeInline: preserved dirty tracking across dma continuations and async readback.
Maxwell macros: preserved gpu owned subranges during page granular cpu uploads.
DiscardWrite: stopped clearing neighboring macro arguments by rounding up ranges.
DMA Step: improved continuation aware dirty sampling.

To the Ender Magnolia crew (maybe 1 or 2 persons): This will fix the dash crash, and the random / shackled beast vaper crashes.

There are some more UE5 issues to go next.

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4473
Reviewed-by: lizzie <lizzie@eden-emu.dev>
Reviewed-by: MaranBr <maranbr@eden-emu.dev>

src/video_core/buffer_cache/buffer_cache.h

@@ -121,7 +121,7 @@ void BufferCache<P>::UnmapGPUMemory(size_t as_id, GPUVAddr gpu_addr, size_t size
 
 template <class P>
 void BufferCache<P>::WriteMemory(DAddr device_addr, u64 size) {
-    if (memory_tracker.IsRegionGpuModified(device_addr, size)) {
+    if (IsRegionGpuModified(device_addr, size)) {
         ClearDownload(device_addr, size);
         gpu_modified_ranges.Subtract(device_addr, size);
     }
@@ -311,11 +311,8 @@ std::pair<typename P::Buffer*, u32> BufferCache<P>::ObtainCPUBuffer(
         MarkWrittenBuffer(buffer_id, device_addr, size);
         break;
     case ObtainBufferOperation::DiscardWrite: {
-        const DAddr device_addr_start = Common::AlignDown(device_addr, 64);
-        const DAddr device_addr_end = Common::AlignUp(device_addr + size, 64);
-        const size_t new_size = device_addr_end - device_addr_start;
-        ClearDownload(device_addr_start, new_size);
-        gpu_modified_ranges.Subtract(device_addr_start, new_size);
+        ClearDownload(device_addr, size);
+        gpu_modified_ranges.Subtract(device_addr, size);
         break;
     }
     default:
@@ -1742,14 +1739,24 @@ bool BufferCache<P>::SynchronizeBuffer(Buffer& buffer, DAddr device_addr, u32 si
     u64 total_size_bytes = 0;
     u64 largest_copy = 0;
     const DAddr buffer_start = buffer.cpu_addr_cached;
-    memory_tracker.ForEachUploadRange(device_addr, size, [&](u64 device_addr_out, u64 range_size) {
+    const auto add_upload = [&](DAddr start, DAddr end) {
+        if (start == end) return;
+        const u64 range_size = end - start;
         upload_copies.push_back(BufferCopy{
             .src_offset = total_size_bytes,
-            .dst_offset = device_addr_out - buffer_start,
+            .dst_offset = start - buffer_start,
             .size = range_size,
         });
         total_size_bytes += range_size;
         largest_copy = (std::max)(largest_copy, range_size);
+    };
+    memory_tracker.ForEachUploadRange(device_addr, size, [&](u64 device_addr_out, u64 range_size) {
+        DAddr upload_start = device_addr_out;
+        gpu_modified_ranges.ForEachInRange(device_addr_out, range_size, [&](DAddr gpu_start, DAddr gpu_end) {
+            add_upload(upload_start, gpu_start);
+            upload_start = gpu_end;
+        });
+        add_upload(upload_start, device_addr_out + range_size);
     });
     if (total_size_bytes == 0) {
         return true;

src/video_core/dma_pusher.cpp

@@ -16,7 +16,7 @@
 namespace Tegra {
 
 constexpr u32 MacroRegistersStart = 0xE00;
-[[maybe_unused]] constexpr u32 ComputeInline = 0x6D;
+constexpr u32 ComputeInline = 0x6D;
 
 DmaPusher::DmaPusher(Core::System& system_, MemoryManager& memory_manager_, Control::ChannelState& channel_state_)
     : system{system_}
@@ -73,11 +73,16 @@ bool DmaPusher::Step() {
         synced = false;
     }
 
-    if (header.size > 0 && dma_state.method >= MacroRegistersStart && subchannels[dma_state.subchannel]) {
-        subchannels[dma_state.subchannel]->current_dirty = memory_manager.IsMemoryDirty(dma_state.dma_get, header.size * sizeof(u32));
-    }
-
     if (header.size > 0) {
+        if (subchannels[dma_state.subchannel] && dma_state.method_count) {
+            const auto engine = subchannel_type[dma_state.subchannel];
+            const bool kepler_payload = engine == Engines::EngineTypes::KeplerCompute && dma_state.method == ComputeInline && dma_state.non_incrementing;
+            const bool macro_payload = engine == Engines::EngineTypes::Maxwell3D && dma_state.method >= MacroRegistersStart;
+            if (kepler_payload || macro_payload) {
+                const size_t words = std::min<size_t>(dma_state.method_count, header.size);
+                subchannels[dma_state.subchannel]->current_dirty = memory_manager.IsMemoryDirty(dma_state.dma_get, words * sizeof(u32));
+            }
+        }
         const bool use_safe = Settings::IsDMALevelDefault() ? Settings::IsGPULevelHigh() : Settings::IsDMALevelSafe();
         if (use_safe) {
             Tegra::Memory::GpuGuestMemory<Tegra::CommandHeader, Tegra::Memory::GuestMemoryFlags::SafeRead>headers(memory_manager, dma_state.dma_get, header.size, &command_headers);

src/video_core/engines/kepler_compute.cpp

@@ -49,13 +49,16 @@ void KeplerCompute::CallMethod(Core::System& system, u32 method, u32 method_argu
     case KEPLER_COMPUTE_REG_INDEX(exec_upload): {
         UploadInfo info{.upload_address = upload_address,
                         .exec_address = upload_state.ExecTargetAddress(),
-                        .copy_size = upload_state.GetUploadSize()};
+                        .copy_size = upload_state.GetUploadSize(),
+                        .was_dirty = upload_dirty};
         uploads.push_back(info);
         upload_state.ProcessExec(regs.exec_upload.linear != 0);
         break;
     }
     case KEPLER_COMPUTE_REG_INDEX(data_upload): {
         upload_address = current_dma_segment;
+        upload_dirty = current_dirty;
+        current_dirty = false;
         upload_state.ProcessData(method_argument, is_last_call);
         break;
     }
@@ -64,9 +67,11 @@ void KeplerCompute::CallMethod(Core::System& system, u32 method, u32 method_argu
 
         for (auto& data : uploads) {
             const GPUVAddr offset = data.exec_address - launch_desc_loc;
-            if (offset / sizeof(u32) == LAUNCH_REG_INDEX(grid_dim_x) &&
-                memory_manager.IsMemoryDirty(data.upload_address, data.copy_size)) {
-                indirect_compute = {data.upload_address};
+            if (offset / sizeof(u32) == LAUNCH_REG_INDEX(grid_dim_x)) {
+                const bool source_dirty = memory_manager.IsMemoryDirty(data.upload_address, data.copy_size);
+                if (data.was_dirty || source_dirty) {
+                    indirect_compute = {data.upload_address};
+                }
             }
         }
         uploads.clear();
@@ -83,6 +88,8 @@ void KeplerCompute::CallMultiMethod(Core::System& system, u32 method, const u32*
     switch (method) {
     case KEPLER_COMPUTE_REG_INDEX(data_upload):
         upload_address = current_dma_segment;
+        upload_dirty = current_dirty;
+        current_dirty = false;
         upload_state.ProcessData(base_start, amount);
         return;
     default:

src/video_core/engines/kepler_compute.h

@@ -226,11 +226,13 @@ private:
     VideoCore::RasterizerInterface* rasterizer = nullptr;
     Upload::State upload_state;
     GPUVAddr upload_address;
+    bool upload_dirty{};
 
     struct UploadInfo {
         GPUVAddr upload_address;
         GPUVAddr exec_address;
         u32 copy_size;
+        bool was_dirty;
     };
     std::vector<UploadInfo> uploads;
     std::optional<GPUVAddr> indirect_compute{};