texture_cache: delete images without walking every channel's tables

Deleting an image stamped every entry of every channel's view-id map
with CORRUPT_ID (validation, which is on in release builds) and cleared
every channel's descriptor table bitmaps on the spot. On Mario Kart
World's title screen (~5,400 entries in 5 channels) that took 362 us
per deleted image, with dozens deleted a frame while the collector runs:
~10% of the time in its frames over 100 ms.

The stamp is debug-only now. The tables are invalidated along with it,
so every entry is written again before it is read. Invalidate() marks a
table and its bitmap is cleared on the next read, once however many
images were deleted in between. Same title screen: 3.7 us per deleted
image.

2 files changed, 19 insertions, 1 deletions

diff --git a/src/video_core/texture_cache/descriptor_table.h b/src/video_core/texture_cache/descriptor_table.h
index e40c128ab5..75383809c5 100644
--- a/src/video_core/texture_cache/descriptor_table.h
+++ b/src/video_core/texture_cache/descriptor_table.h
@@ -30,11 +30,17 @@ public:
     }
 
     void Invalidate() noexcept {
-        std::ranges::fill(read_descriptors, 0);
+        // Cleared on the next read: the texture cache invalidates every channel's tables for
+        // each image it deletes, dozens of times a frame while its collector runs.
+        invalidated = true;
     }
 
     [[nodiscard]] std::pair<T, bool> Read(Tegra::MemoryManager const& gpu_memory, u32 index) noexcept {
         DEBUG_ASSERT(index <= current_limit);
+        if (invalidated) [[unlikely]] {
+            std::ranges::fill(read_descriptors, 0);
+            invalidated = false;
+        }
         const GPUVAddr gpu_addr = current_gpu_addr + index * sizeof(T);
         std::pair<T, bool> result;
         gpu_memory.ReadBlockUnsafe(gpu_addr, std::addressof(result.first), sizeof(T));
@@ -60,6 +66,7 @@ public:
         read_descriptors.resize(Common::DivCeil(num_descriptors, 64U));
         old_size = (std::min)(old_size, read_descriptors.size());
         std::fill(read_descriptors.begin(), read_descriptors.begin() + old_size, 0);
+        invalidated = false;
         //
         descriptors.resize(num_descriptors);
     }
@@ -68,6 +75,7 @@ public:
     std::vector<T> descriptors;
     GPUVAddr current_gpu_addr{};
     u32 current_limit{};
+    bool invalidated{}; ///< read_descriptors is to be cleared before the next read
 };
 
 } // namespace VideoCommon
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index c8aaf8dff5..559423dbee 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -1712,9 +1712,14 @@ void TextureCache<P>::InvalidateScale(Image& image) {
     for (size_t c : active_channel_ids) {
         auto& channel_info = channel_storage[c];
 
+#ifndef NDEBUG
+        // A walk over every entry (title 1: ~5,400 in 5 channels) for each deleted image: debug
+        // builds only. The tables are invalidated below, so each entry is written again before
+        // it is read.
         if constexpr (ENABLE_VALIDATION)
             for (auto& e : channel_info.image_view_ids)
                 e.second = CORRUPT_ID;
+#endif
         channel_info.graphics_image_table.Invalidate();
         channel_info.compute_image_table.Invalidate();
     }
@@ -2906,9 +2911,14 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) {
     }
     for (size_t c : active_channel_ids) {
         auto& channel_info = channel_storage[c];
+#ifndef NDEBUG
+        // A walk over every entry (title 1: ~5,400 in 5 channels) for each deleted image: debug
+        // builds only. The tables are invalidated below, so each entry is written again before
+        // it is read.
         if constexpr (ENABLE_VALIDATION)
             for (auto& e : channel_info.image_view_ids)
                 e.second = CORRUPT_ID;
+#endif
         channel_info.graphics_image_table.Invalidate();
         channel_info.compute_image_table.Invalidate();
     }