[D3D12] SHM functions needed for resolves

src/xenia/gpu/d3d12/shared_memory.cc

@@ -193,8 +193,7 @@ void SharedMemory::UnwatchMemoryRange(WatchHandle handle) {
   UnlinkWatchRange(reinterpret_cast<WatchRange*>(handle));
 }
 
-bool SharedMemory::RequestRange(uint32_t start, uint32_t length,
-                                ID3D12GraphicsCommandList* command_list) {
+bool SharedMemory::MakeTilesResident(uint32_t start, uint32_t length) {
   if (length == 0) {
     // Some texture is empty, for example - safe to draw in this case.
     return true;
@@ -204,16 +203,13 @@ bool SharedMemory::RequestRange(uint32_t start, uint32_t length,
     // Exceeds the physical address space.
     return false;
   }
-  uint32_t last = start + length - 1;
 
-#if FINE_GRAINED_DRAW_SCOPES
-  SCOPE_profile_cpu_f("gpu");
-#endif  // FINE_GRAINED_DRAW_SCOPES
+  if (!FLAGS_d3d12_tiled_resources) {
+    return true;
+  }
 
-  // Ensure all tile heaps are present.
-  if (FLAGS_d3d12_tiled_resources) {
   uint32_t heap_first = start >> kHeapSizeLog2;
-    uint32_t heap_last = last >> kHeapSizeLog2;
+  uint32_t heap_last = (start + length - 1) >> kHeapSizeLog2;
   for (uint32_t i = heap_first; i <= heap_last; ++i) {
     if (heaps_[i] != nullptr) {
       continue;
@@ -255,6 +251,29 @@ bool SharedMemory::RequestRange(uint32_t start, uint32_t length,
         &range_flags, &heap_range_start_offset, &range_tile_count,
         D3D12_TILE_MAPPING_FLAG_NONE);
   }
+  return true;
+}
+
+bool SharedMemory::RequestRange(uint32_t start, uint32_t length,
+                                ID3D12GraphicsCommandList* command_list) {
+  if (length == 0) {
+    // Some texture is empty, for example - safe to draw in this case.
+    return true;
+  }
+  start &= kAddressMask;
+  if (start >= kBufferSize || (kBufferSize - start) < length) {
+    // Exceeds the physical address space.
+    return false;
+  }
+  uint32_t last = start + length - 1;
+
+#if FINE_GRAINED_DRAW_SCOPES
+  SCOPE_profile_cpu_f("gpu");
+#endif  // FINE_GRAINED_DRAW_SCOPES
+
+  // Ensure all tile heaps are present.
+  if (!MakeTilesResident(start, length)) {
+    return false;
   }
 
   // Upload and protect used ranges.
@@ -296,6 +315,42 @@ bool SharedMemory::RequestRange(uint32_t start, uint32_t length,
   return true;
 }
 
+void SharedMemory::RangeWrittenByGPU(uint32_t start, uint32_t length) {
+  start &= kAddressMask;
+  if (length == 0 || start >= kBufferSize) {
+    return;
+  }
+  length = std::min(length, kBufferSize - start);
+  uint32_t end = start + length - 1;
+  uint32_t page_first = start >> page_size_log2_;
+  uint32_t page_last = end >> page_size_log2_;
+  uint32_t bucket_first = start >> kWatchBucketSizeLog2;
+  uint32_t bucket_last = end >> kWatchBucketSizeLog2;
+
+  std::lock_guard<std::recursive_mutex> lock(validity_mutex_);
+
+  // Trigger modification callbacks so, for instance, resolved data is loaded to
+  // the texture.
+  for (uint32_t i = bucket_first; i <= bucket_last; ++i) {
+    WatchNode* node = watch_buckets_[i];
+    while (node != nullptr) {
+      WatchRange* range = node->range;
+      // Store the next node now since when the callback is triggered, the links
+      // will be broken.
+      node = node->bucket_node_next;
+      if (page_first <= range->page_last && page_last >= range->page_first) {
+        range->callback(range->callback_context, range->callback_data,
+                        range->callback_argument);
+        UnlinkWatchRange(range);
+      }
+    }
+  }
+
+  // Mark the range as valid (so pages are not reuploaded until modified by the
+  // CPU) and protect it so the CPU can reuse it.
+  MakeRangeValid(page_first, page_last - page_first + 1);
+}
+
 void SharedMemory::MakeRangeValid(uint32_t valid_page_first,
                                   uint32_t valid_page_count) {
   if (valid_page_first >= page_count_ || valid_page_count == 0) {

src/xenia/gpu/d3d12/shared_memory.h

@@ -67,6 +67,10 @@ class SharedMemory {
   inline void LockWatchMutex() { validity_mutex_.lock(); }
   inline void UnlockWatchMutex() { validity_mutex_.unlock(); }
 
+  // Ensures the buffer tiles backing the range are resident, but doesn't upload
+  // anything.
+  bool MakeTilesResident(uint32_t start, uint32_t length);
+
   // Checks if the range has been updated, uploads new data if needed and
   // ensures the buffer tiles backing the range are resident. May transition the
   // tiled buffer to copy destination - call this before UseForReading or
@@ -75,6 +79,12 @@ class SharedMemory {
   bool RequestRange(uint32_t start, uint32_t length,
                     ID3D12GraphicsCommandList* command_list);
 
+  // Marks the range as containing GPU-generated data (such as resolves),
+  // triggering modification callbacks, making it valid (so pages are not
+  // copied from the main memory until they're modified by the CPU) and
+  // protecting it.
+  void RangeWrittenByGPU(uint32_t start, uint32_t length);
+
   // Makes the buffer usable for vertices, indices and texture untiling.
   void UseForReading(ID3D12GraphicsCommandList* command_list);
   // Makes the buffer usable for texture tiling after a resolve.