early-access version 1877

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 1875.
+This is the source code for early-access 1877.
 
 ## Legal Notice
 

src/video_core/buffer_cache/buffer_cache.h

@@ -550,6 +550,10 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
     auto& src_buffer = slot_buffers[buffer_a];
     auto& dest_buffer = slot_buffers[buffer_b];
     SynchronizeBuffer(src_buffer, *cpu_src_address, static_cast<u32>(amount));
+    const VAddr aligned_dst = Common::AlignUp(*cpu_dest_address, 64);
+    const u64 align_diff = aligned_dst - *cpu_dest_address;
+    const u64 new_amount = align_diff > amount ? 0 : amount - align_diff;
+    dest_buffer.UnmarkRegionAsCpuModified(aligned_dst, Common::AlignDown(new_amount, 64));
     SynchronizeBuffer(dest_buffer, *cpu_dest_address, static_cast<u32>(amount));
     std::array copies{BufferCopy{
         .src_offset = src_buffer.Offset(*cpu_src_address),
@@ -569,12 +573,13 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
     ForEachWrittenRange(*cpu_src_address, amount, mirror);
     // This subtraction in this order is important for overlapping copies.
     common_ranges.subtract(subtract_interval);
+    bool atleast_1_download = tmp_intervals.size() != 0;
     for (const IntervalType add_interval : tmp_intervals) {
         common_ranges.add(add_interval);
     }
 
     runtime.CopyBuffer(dest_buffer, src_buffer, copies);
-    if (IsRegionGpuModified(*cpu_src_address, amount)) {
+    if (atleast_1_download) {
         dest_buffer.MarkRegionAsGpuModified(*cpu_dest_address, amount);
     }
     std::vector<u8> tmp_buffer(amount);

src/video_core/engines/maxwell_dma.cpp

@@ -182,8 +182,13 @@ void MaxwellDMA::CopyPitchToBlockLinear() {
         write_buffer.resize(dst_size);
     }
 
-    memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
+    if (Settings::IsGPULevelExtreme()) {
-    memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
+        memory_manager.ReadBlock(regs.offset_in, read_buffer.data(), src_size);
+        memory_manager.ReadBlock(regs.offset_out, write_buffer.data(), dst_size);
+    } else {
+        memory_manager.ReadBlockUnsafe(regs.offset_in, read_buffer.data(), src_size);
+        memory_manager.ReadBlockUnsafe(regs.offset_out, write_buffer.data(), dst_size);
+    }
 
     // If the input is linear and the output is tiled, swizzle the input and copy it over.
     if (regs.dst_params.block_size.depth > 0) {

src/video_core/memory_manager.cpp

@@ -345,7 +345,7 @@ void MemoryManager::WriteBlock(GPUVAddr gpu_dest_addr, const void* src_buffer, s
 
             // Invalidate must happen on the rasterizer interface, such that memory is always
             // synchronous when it is written (even when in asynchronous GPU mode).
-            rasterizer->InvalidateRegion(dest_addr, copy_amount);
+            rasterizer->UnmapMemory(dest_addr, copy_amount);
             system.Memory().WriteBlockUnsafe(dest_addr, src_buffer, copy_amount);
         }
 

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -222,13 +222,11 @@ void RasterizerVulkan::Clear() {
         .height = std::min(clear_rect.rect.extent.height, render_area.height),
     };
 
-    const u32 color_attachment = regs.clear_buffers.RT;
+    if (use_color) {
-    const auto attachment_aspect_mask = framebuffer->ImageRanges()[color_attachment].aspectMask;
-    const bool is_color_rt = (attachment_aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT) != 0;
-    if (use_color && is_color_rt) {
         VkClearValue clear_value;
         std::memcpy(clear_value.color.float32, regs.clear_color, sizeof(regs.clear_color));
 
+        const u32 color_attachment = regs.clear_buffers.RT;
         scheduler.Record([color_attachment, clear_value, clear_rect](vk::CommandBuffer cmdbuf) {
             const VkClearAttachment attachment{
                 .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,