Merge remote-tracking branch 'origin/fix/tc-count-sentenced' into test/linux-all-prs

2 files changed, 41 insertions, 1 deletions

diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 44626318d1..d7990fb24e 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -255,9 +255,23 @@ void TextureCache<P>::TickFrame() {
             }
         }
     }
+    if constexpr (requires { runtime.CurrentTick(); }) {
+        while (!gpu_sentenced_images.empty() &&
+               runtime.IsFree(gpu_sentenced_images.front().tick)) {
+            gpu_sentenced_bytes -= gpu_sentenced_images.front().bytes;
+            gpu_sentenced_images.pop_front();
+        }
+    }
     // If we can obtain the memory info, use it instead of the estimate.
     if (runtime.CanReportMemoryUsage()) {
         total_used_memory = runtime.GetDeviceMemoryUsage();
+        // Images the collector deleted keep their memory for TICKS_TO_DESTROY frames: counting
+        // it kept the collector in aggressive mode, deleting 40+ images a frame for frames after
+        // it had freed enough (title 1: frames over 100 ms for seconds at a time).
+        for (const u64 bytes : sentenced_image_bytes) {
+            total_used_memory -= (std::min)(total_used_memory, bytes);
+        }
+        total_used_memory -= (std::min)(total_used_memory, gpu_sentenced_bytes);
     }
     // Every 600 frames: device memory in use against the collector's thresholds, and what the
     // images themselves account for (VRAM pressure makes the collector thrash).
@@ -324,6 +338,8 @@ void TextureCache<P>::TickFrame() {
         RunGarbageCollector();
     }
     sentenced_images.Tick();
+    sentenced_image_index = (sentenced_image_index + 1) % TICKS_TO_DESTROY;
+    sentenced_image_bytes[sentenced_image_index] = 0;
     sentenced_framebuffers.Tick();
     sentenced_image_view.Tick();
     TickAsyncDecode();
@@ -2939,6 +2955,8 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) {
     }
     // TickFrame sets the total from the driver, which can be under the sum of these estimates.
     total_used_memory -= (std::min)(total_used_memory, ImageMemoryBytes(slot_images[image_id]));
+    const u64 freed_bytes = (image.HasScaled() ? GetScaledImageSizeBytes(image) : 0) +
+                            ImageMemoryBytes(slot_images[image_id]);
     const GPUVAddr gpu_addr = image.gpu_addr;
     const auto alloc_it = image_allocs_table.find(gpu_addr);
     if (alloc_it == image_allocs_table.end()) {
@@ -3003,7 +3021,14 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) {
     if (retained) {
         LOG_INFO(Render_Vulkan, "Retained image for queued download: id={}", image_id.Value());
     } else if (!immediate_delete) {
-        sentenced_images.Push(std::move(slot_images[image_id]));
+        if constexpr (requires { runtime.CurrentTick(); }) {
+            gpu_sentenced_images.push_back(
+                {runtime.CurrentTick(), freed_bytes, std::move(slot_images[image_id])});
+            gpu_sentenced_bytes += freed_bytes;
+        } else {
+            sentenced_images.Push(std::move(slot_images[image_id]));
+            sentenced_image_bytes[sentenced_image_index] += freed_bytes;
+        }
     }
     slot_images.erase(image_id);
 
diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h
index 1f52909914..99de167513 100644
--- a/src/video_core/texture_cache/texture_cache_base.h
+++ b/src/video_core/texture_cache/texture_cache_base.h
@@ -567,6 +567,21 @@ private:
     static constexpr size_t TICKS_TO_DESTROY = 8;
 #endif
     DelayedDestructionRing<Image, TICKS_TO_DESTROY> sentenced_images;
+    /// Device memory of the images in each slot of sentenced_images (same index): deleted, but
+    /// counted in the driver's usage until the ring destroys them.
+    std::array<u64, TICKS_TO_DESTROY> sentenced_image_bytes{};
+    size_t sentenced_image_index = 0;
+    /// Deleted images on runtimes that report GPU progress (CurrentTick, IsFree): destroyed, with
+    /// their device memory, as soon as the GPU is past them rather than TICKS_TO_DESTROY frames
+    /// later. While the collector deletes dozens a frame, eight frames of them held 1-1.5 GB of
+    /// VRAM above its threshold (title 1 on an 8 GB card: 6.6-7.0 GB in use against 5 GiB).
+    struct GpuSentencedImage {
+        u64 tick;
+        u64 bytes;
+        Image image;
+    };
+    std::deque<GpuSentencedImage> gpu_sentenced_images;
+    u64 gpu_sentenced_bytes{};
     DelayedDestructionRing<ImageView, TICKS_TO_DESTROY> sentenced_image_view;
     DelayedDestructionRing<Framebuffer, TICKS_TO_DESTROY> sentenced_framebuffers;