texture_cache: Basic handling of partially resident images (#3066)

* texture_cache: Avoid gpu tracking assert on sparse image

At the moment just take the easy way of creating the entire image normally and uploading unmapped subresources are zero

* tile_manager: Downgrade assert to error

* fix macos

src/core/libraries/kernel/memory.cpp

@@ -660,6 +660,9 @@ int PS4_SYSV_ABI sceKernelSetPrtAperture(int id, VAddr address, size_t size) {
                 "PRT aperture id = {}, address = {:#x}, size = {:#x} is set but not used", id,
                 address, size);
 
+    auto* memory = Core::Memory::Instance();
+    memory->SetPrtArea(id, address, size);
+
     PrtApertures[id] = {address, size};
     return ORBIS_OK;
 }

src/core/memory.cpp

@@ -95,6 +95,46 @@ u64 MemoryManager::ClampRangeSize(VAddr virtual_addr, u64 size) {
     return clamped_size;
 }
 
+void MemoryManager::SetPrtArea(u32 id, VAddr address, u64 size) {
+    PrtArea& area = prt_areas[id];
+    if (area.mapped) {
+        rasterizer->UnmapMemory(area.start, area.end - area.start);
+    }
+
+    area.start = address;
+    area.end = address + size;
+    area.mapped = true;
+
+    // Pretend the entire PRT area is mapped to avoid GPU tracking errors.
+    // The caches will use CopySparseMemory to fetch data which avoids unmapped areas.
+    rasterizer->MapMemory(address, size);
+}
+
+void MemoryManager::CopySparseMemory(VAddr virtual_addr, u8* dest, u64 size) {
+    const bool is_sparse = std::ranges::any_of(
+        prt_areas, [&](const PrtArea& area) { return area.Overlaps(virtual_addr, size); });
+    if (!is_sparse) {
+        std::memcpy(dest, std::bit_cast<const u8*>(virtual_addr), size);
+        return;
+    }
+
+    auto vma = FindVMA(virtual_addr);
+    ASSERT_MSG(vma->second.Contains(virtual_addr, 0),
+               "Attempted to access invalid GPU address {:#x}", virtual_addr);
+    while (size) {
+        u64 copy_size = std::min<u64>(vma->second.size - (virtual_addr - vma->first), size);
+        if (vma->second.IsFree()) {
+            std::memset(dest, 0, copy_size);
+        } else {
+            std::memcpy(dest, std::bit_cast<const u8*>(virtual_addr), copy_size);
+        }
+        size -= copy_size;
+        virtual_addr += copy_size;
+        dest += copy_size;
+        ++vma;
+    }
+}
+
 bool MemoryManager::TryWriteBacking(void* address, const void* data, u32 num_bytes) {
     const VAddr virtual_addr = std::bit_cast<VAddr>(address);
     const auto& vma = FindVMA(virtual_addr)->second;

src/core/memory.h

@@ -172,6 +172,10 @@ public:
 
     u64 ClampRangeSize(VAddr virtual_addr, u64 size);
 
+    void SetPrtArea(u32 id, VAddr address, u64 size);
+
+    void CopySparseMemory(VAddr source, u8* dest, u64 size);
+
     bool TryWriteBacking(void* address, const void* data, u32 num_bytes);
 
     void SetupMemoryRegions(u64 flexible_size, bool use_extended_mem1, bool use_extended_mem2);
@@ -275,6 +279,18 @@ private:
     size_t pool_budget{};
     Vulkan::Rasterizer* rasterizer{};
 
+    struct PrtArea {
+        VAddr start;
+        VAddr end;
+        bool mapped;
+
+        bool Overlaps(VAddr test_address, u64 test_size) const {
+            const VAddr overlap_end = test_address + test_size;
+            return start < overlap_end && test_address < end;
+        }
+    };
+    std::array<PrtArea, 3> prt_areas{};
+
     friend class ::Core::Devtools::Widget::MemoryMapViewer;
 };
 

src/video_core/buffer_cache/buffer_cache.cpp

@@ -6,6 +6,7 @@
 #include "common/debug.h"
 #include "common/scope_exit.h"
 #include "common/types.h"
+#include "core/memory.h"
 #include "video_core/amdgpu/liverpool.h"
 #include "video_core/buffer_cache/buffer_cache.h"
 #include "video_core/host_shaders/fault_buffer_process_comp.h"
@@ -28,7 +29,7 @@ BufferCache::BufferCache(const Vulkan::Instance& instance_, Vulkan::Scheduler& s
                          Vulkan::Rasterizer& rasterizer_, AmdGpu::Liverpool* liverpool_,
                          TextureCache& texture_cache_, PageManager& tracker_)
     : instance{instance_}, scheduler{scheduler_}, rasterizer{rasterizer_}, liverpool{liverpool_},
-      texture_cache{texture_cache_}, tracker{tracker_},
+      memory{Core::Memory::Instance()}, texture_cache{texture_cache_}, tracker{tracker_},
       staging_buffer{instance, scheduler, MemoryUsage::Upload, StagingBufferSize},
       stream_buffer{instance, scheduler, MemoryUsage::Stream, UboStreamBufferSize},
       download_buffer(instance, scheduler, MemoryUsage::Download, DownloadBufferSize),
@@ -365,7 +366,9 @@ std::pair<Buffer*, u32> BufferCache::ObtainViewBuffer(VAddr gpu_addr, u32 size,
         return ObtainBuffer(gpu_addr, size, false, false);
     }
     // In all other cases, just do a CPU copy to the staging buffer.
-    const u32 offset = staging_buffer.Copy(gpu_addr, size, 16);
+    const auto [data, offset] = staging_buffer.Map(size, 16);
+    memory->CopySparseMemory(gpu_addr, data, size);
+    staging_buffer.Commit();
     return {&staging_buffer, offset};
 }
 

src/video_core/buffer_cache/buffer_cache.h

@@ -17,6 +17,10 @@ namespace AmdGpu {
 struct Liverpool;
 }
 
+namespace Core {
+class MemoryManager;
+}
+
 namespace Shader {
 namespace Gcn {
 struct FetchShaderData;
@@ -183,6 +187,7 @@ private:
     Vulkan::Scheduler& scheduler;
     Vulkan::Rasterizer& rasterizer;
     AmdGpu::Liverpool* liverpool;
+    Core::MemoryManager* memory;
     TextureCache& texture_cache;
     PageManager& tracker;
     StreamBuffer staging_buffer;