Initial Reaper Setup

WIP

src/video_core/texture_cache/image_base.cpp

@@ -113,6 +113,23 @@ void ImageBase::InsertView(const ImageViewInfo& view_info, ImageViewId image_vie
     image_view_ids.push_back(image_view_id);
 }
 
+bool ImageBase::IsSafeDownload() const noexcept {
+    // Skip images that were not modified from the GPU
+    if (False(flags & ImageFlagBits::GpuModified)) {
+        return false;
+    }
+    // Skip images that .are. modified from the CPU
+    // We don't want to write sensitive data from the guest
+    if (True(flags & ImageFlagBits::CpuModified)) {
+        return false;
+    }
+    if (info.num_samples > 1) {
+        LOG_WARNING(HW_GPU, "MSAA image downloads are not implemented");
+        return false;
+    }
+    return true;
+}
+
 void AddImageAlias(ImageBase& lhs, ImageBase& rhs, ImageId lhs_id, ImageId rhs_id) {
     static constexpr auto OPTIONS = RelaxedOptions::Size | RelaxedOptions::Format;
     ASSERT(lhs.info.type == rhs.info.type);

src/video_core/texture_cache/image_base.h

@@ -44,6 +44,8 @@ struct ImageBase {
 
     void InsertView(const ImageViewInfo& view_info, ImageViewId image_view_id);
 
+    [[nodiscard]] bool IsSafeDownload() const noexcept;
+
     [[nodiscard]] bool Overlaps(VAddr overlap_cpu_addr, size_t overlap_size) const noexcept {
         const VAddr overlap_end = overlap_cpu_addr + overlap_size;
         return cpu_addr < overlap_end && overlap_cpu_addr < cpu_addr_end;

src/video_core/texture_cache/slot_vector.h

@@ -5,6 +5,7 @@
 #pragma once
 
 #include <array>
+#include <bit>
 #include <concepts>
 #include <numeric>
 #include <type_traits>
@@ -32,6 +33,60 @@ template <class T>
 requires std::is_nothrow_move_assignable_v<T>&&
     std::is_nothrow_move_constructible_v<T> class SlotVector {
 public:
+    class Iterator {
+        friend SlotVector<T>;
+
+    public:
+        constexpr Iterator() = default;
+
+        Iterator& operator++() noexcept {
+            const u64* const bitset = slot_vector->stored_bitset.data();
+            const u32 size = static_cast<u32>(slot_vector->stored_bitset.size()) * 64;
+            if (id.index < size) {
+                do {
+                    ++id.index;
+                } while (id.index < size && !IsValid(bitset));
+                if (id.index == size) {
+                    id.index = SlotId::INVALID_INDEX;
+                }
+            }
+            return *this;
+        }
+
+        Iterator operator++(int) noexcept {
+            const Iterator copy{*this};
+            ++*this;
+            return copy;
+        }
+
+        bool operator==(const Iterator& other) const noexcept {
+            return id.index == other.id.index;
+        }
+
+        bool operator!=(const Iterator& other) const noexcept {
+            return id.index != other.id.index;
+        }
+
+        std::pair<SlotId, T*> operator*() const noexcept {
+            return {id, std::addressof((*slot_vector)[id])};
+        }
+
+        T* operator->() const noexcept {
+            return std::addressof((*slot_vector)[id]);
+        }
+
+    private:
+        Iterator(SlotVector<T>* slot_vector_, SlotId id_) noexcept
+            : slot_vector{slot_vector_}, id{id_} {}
+
+        bool IsValid(const u64* bitset) noexcept {
+            return ((bitset[id.index / 64] >> (id.index % 64)) & 1) != 0;
+        }
+
+        SlotVector<T>* slot_vector;
+        SlotId id;
+    };
+
     ~SlotVector() noexcept {
         size_t index = 0;
         for (u64 bits : stored_bitset) {
@@ -70,6 +125,20 @@ public:
         ResetStorageBit(id.index);
     }
 
+    [[nodiscard]] Iterator begin() noexcept {
+        const auto it = std::ranges::find_if(stored_bitset, [](u64 value) { return value != 0; });
+        if (it == stored_bitset.end()) {
+            return end();
+        }
+        const u32 word_index = static_cast<u32>(std::distance(it, stored_bitset.begin()));
+        const SlotId first_id{word_index * 64 + static_cast<u32>(std::countr_zero(*it))};
+        return Iterator(this, first_id);
+    }
+
+    [[nodiscard]] Iterator end() noexcept {
+        return Iterator(this, SlotId{SlotId::INVALID_INDEX});
+    }
+
 private:
     struct NonTrivialDummy {
         NonTrivialDummy() noexcept {}
@@ -140,7 +209,6 @@ private:
 
     Entry* values = nullptr;
     size_t values_capacity = 0;
-    size_t values_size = 0;
 
     std::vector<u64> stored_bitset;
     std::vector<u32> free_list;

src/video_core/texture_cache/texture_cache.h

@@ -353,6 +353,7 @@ private:
 
     u64 modification_tick = 0;
     u64 frame_tick = 0;
+    typename SlotVector<Image>::Iterator deletion_iterator;
 };
 
 template <class P>
@@ -373,10 +374,41 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface&
     // This way the null resource becomes a compile time constant
     void(slot_image_views.insert(runtime, NullImageParams{}));
     void(slot_samplers.insert(runtime, sampler_descriptor));
+
+    deletion_iterator = slot_images.begin();
 }
 
 template <class P>
 void TextureCache<P>::TickFrame() {
+    static constexpr u64 ticks_to_destroy = 120;
+    int num_iterations = 32;
+    for (; num_iterations > 0; --num_iterations) {
+        if (deletion_iterator == slot_images.end()) {
+            deletion_iterator = slot_images.begin();
+            if (deletion_iterator == slot_images.end()) {
+                break;
+            }
+        }
+        const auto [image_id, image] = *deletion_iterator;
+        if (image->frame_tick + ticks_to_destroy < frame_tick) {
+            if (image->IsSafeDownload() &&
+                std::ranges::none_of(image->aliased_images, [&](const AliasedImage& alias) {
+                    return slot_images[alias.id].modification_tick > image->modification_tick;
+                })) {
+                auto map = runtime.DownloadStagingBuffer(image->unswizzled_size_bytes);
+                const auto copies = FullDownloadCopies(image->info);
+                image->DownloadMemory(map, copies);
+                runtime.Finish();
+                SwizzleImage(gpu_memory, image->gpu_addr, image->info, copies, map.mapped_span);
+            }
+            if (True(image->flags & ImageFlagBits::Tracked)) {
+                UntrackImage(*image);
+            }
+            UnregisterImage(image_id);
+            DeleteImage(image_id);
+        }
+        ++deletion_iterator;
+    }
     // Tick sentenced resources in this order to ensure they are destroyed in the right order
     sentenced_images.Tick();
     sentenced_framebuffers.Tick();
@@ -568,17 +600,7 @@ template <class P>
 void TextureCache<P>::DownloadMemory(VAddr cpu_addr, size_t size) {
     std::vector<ImageId> images;
     ForEachImageInRegion(cpu_addr, size, [this, &images](ImageId image_id, ImageBase& image) {
-        // Skip images that were not modified from the GPU
-        if (False(image.flags & ImageFlagBits::GpuModified)) {
-            return;
-        }
-        // Skip images that .are. modified from the CPU
-        // We don't want to write sensitive data from the guest
-        if (True(image.flags & ImageFlagBits::CpuModified)) {
-            return;
-        }
-        if (image.info.num_samples > 1) {
-            LOG_WARNING(HW_GPU, "MSAA image downloads are not implemented");
+        if (!image.IsSafeDownload()) {
             return;
         }
         image.flags &= ~ImageFlagBits::GpuModified;