From 47d0d292d5cc5f0404e126023279db7decd532ac Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow
Date: Wed, 28 Jun 2023 06:28:13 +0200
Subject: MemoryTracking: Initial setup of atomic writes.

---
 src/video_core/buffer_cache/buffer_cache.h      | 26 ++++++++++++++++++++++---
 src/video_core/buffer_cache/buffer_cache_base.h |  2 ++
 2 files changed, 25 insertions(+), 3 deletions(-)

(limited to 'src/video_core/buffer_cache')

diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 58a45ab67..9239ad862 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -115,7 +115,21 @@ void BufferCache<P>::WriteMemory(VAddr cpu_addr, u64 size) {
 
 template <class P>
 void BufferCache<P>::CachedWriteMemory(VAddr cpu_addr, u64 size) {
-    memory_tracker.CachedCpuWrite(cpu_addr, size);
+    const bool is_dirty = IsRegionRegistered(cpu_addr, size);
+    if (!is_dirty) {
+        return;
+    }
+    VAddr aligned_start = Common::AlignDown(cpu_addr, YUZU_PAGESIZE);
+    VAddr aligned_end = Common::AlignUp(cpu_addr + size, YUZU_PAGESIZE);
+    if (!IsRegionGpuModified(aligned_start, aligned_end - aligned_start)) {
+        WriteMemory(cpu_addr, size);
+        return;
+    }
+
+    tmp_buffer.resize_destructive(size);
+    cpu_memory.ReadBlockUnsafe(cpu_addr, tmp_buffer.data(), size);
+
+    InlineMemoryImplementation(cpu_addr, size, tmp_buffer);
 }
 
 template <class P>
@@ -1553,6 +1567,14 @@ bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size,
         return false;
     }
 
+    InlineMemoryImplementation(dest_address, copy_size, inlined_buffer);
+
+    return true;
+}
+
+template <class P>
+void BufferCache<P>::InlineMemoryImplementation(VAddr dest_address, size_t copy_size,
+                                  std::span<const u8> inlined_buffer) {
     const IntervalType subtract_interval{dest_address, dest_address + copy_size};
     ClearDownload(subtract_interval);
     common_ranges.subtract(subtract_interval);
@@ -1574,8 +1596,6 @@ bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size,
     } else {
         buffer.ImmediateUpload(buffer.Offset(dest_address), inlined_buffer.first(copy_size));
     }
-
-    return true;
 }
 
 template <class P>
diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h
index fe6068cfe..4d9bab7f7 100644
--- a/src/video_core/buffer_cache/buffer_cache_base.h
+++ b/src/video_core/buffer_cache/buffer_cache_base.h
@@ -543,6 +543,8 @@ private:
 
     void ClearDownload(IntervalType subtract_interval);
 
+    void InlineMemoryImplementation(VAddr dest_address, size_t copy_size, std::span<const u8> inlined_buffer);
+
     VideoCore::RasterizerInterface& rasterizer;
     Core::Memory::Memory& cpu_memory;
 
-- 
cgit v1.2.3


From da440da9f54cc860f3c69da685a415d5ec9d7b64 Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow
Date: Wed, 28 Jun 2023 19:32:50 +0200
Subject: Memory Tracking: Optimize tracking to only use atomic writes when
 contested with the host GPU

---
 src/video_core/buffer_cache/buffer_cache.h      | 15 ++++++++++++++-
 src/video_core/buffer_cache/buffer_cache_base.h |  5 ++++-
 2 files changed, 18 insertions(+), 2 deletions(-)

(limited to 'src/video_core/buffer_cache')

diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 9239ad862..b5ed3380f 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -132,6 +132,19 @@ void BufferCache<P>::CachedWriteMemory(VAddr cpu_addr, u64 size) {
     InlineMemoryImplementation(cpu_addr, size, tmp_buffer);
 }
 
+template <class P>
+bool BufferCache<P>::OnCPUWrite(VAddr cpu_addr, u64 size) {
+    const bool is_dirty = IsRegionRegistered(cpu_addr, size);
+    if (!is_dirty) {
+        return false;
+    }
+    if (memory_tracker.IsRegionGpuModified(cpu_addr, size)) {
+        return true;
+    }
+    WriteMemory(cpu_addr, size);
+    return false;
+}
+
 template <class P>
 std::optional<VideoCore::RasterizerDownloadArea> BufferCache<P>::GetFlushArea(VAddr cpu_addr,
                                                                               u64 size) {
@@ -1574,7 +1587,7 @@ bool BufferCache<P>::InlineMemory(VAddr dest_address, size_t copy_size,
 
 template <class P>
 void BufferCache<P>::InlineMemoryImplementation(VAddr dest_address, size_t copy_size,
-                                  std::span<const u8> inlined_buffer) {
+                                                std::span<const u8> inlined_buffer) {
     const IntervalType subtract_interval{dest_address, dest_address + copy_size};
     ClearDownload(subtract_interval);
     common_ranges.subtract(subtract_interval);
diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h
index 4d9bab7f7..460fc7551 100644
--- a/src/video_core/buffer_cache/buffer_cache_base.h
+++ b/src/video_core/buffer_cache/buffer_cache_base.h
@@ -245,6 +245,8 @@ public:
 
     void CachedWriteMemory(VAddr cpu_addr, u64 size);
 
+    bool OnCPUWrite(VAddr cpu_addr, u64 size);
+
     void DownloadMemory(VAddr cpu_addr, u64 size);
 
     std::optional<VideoCore::RasterizerDownloadArea> GetFlushArea(VAddr cpu_addr, u64 size);
@@ -543,7 +545,8 @@ private:
 
     void ClearDownload(IntervalType subtract_interval);
 
-    void InlineMemoryImplementation(VAddr dest_address, size_t copy_size, std::span<const u8> inlined_buffer);
+    void InlineMemoryImplementation(VAddr dest_address, size_t copy_size,
+                                    std::span<const u8> inlined_buffer);
 
     VideoCore::RasterizerInterface& rasterizer;
     Core::Memory::Memory& cpu_memory;
-- 
cgit v1.2.3