Rasterizer refactor hotfixes (#6465)

* texture_codec: Clamp buffer end to tiled buffer size

* Fixes crash on Pokemon Super Mystery Dungeon

* rasterizer_cache: Use rect for duplicate surface

* Fixes broken bloom in fire emblem

* surface_params: Check levels for exact match

* It was removed previously to prevent copies when games used the base level of a multi level surface. FE on the other hand will first use the base level and then use it as a face of a cubemap with many levels. So instead check if the surface equal or more levels and consider it an exact match in that case

* gl_texture_runtime: Bind old tex to 2D target

* Fixes a small error opengl would print when creating texture cubes

* gl_blit_helper: Fix nearest filter

* Use texture unit 2 which has the nearest sampler bound

src/video_core/rasterizer_cache/rasterizer_cache.cpp

@@ -725,8 +725,8 @@ void RasterizerCache::DuplicateSurface(const SurfaceRef& src_surface,
     const TextureCopy copy = {
         .src_level = 0,
         .dst_level = 0,
-        .src_offset = {0, 0},
-        .dst_offset = {0, 0},
+        .src_offset = {src_rect.left, src_rect.bottom},
+        .dst_offset = {dst_rect.left, dst_rect.bottom},
         .extent = {src_rect.GetWidth(), src_rect.GetHeight()},
     };
     runtime.CopyTextures(*src_surface, *dest_surface, copy);

src/video_core/rasterizer_cache/surface_params.cpp

@@ -11,7 +11,7 @@ bool SurfaceParams::ExactMatch(const SurfaceParams& other_surface) const {
     return std::tie(other_surface.addr, other_surface.width, other_surface.height,
                     other_surface.stride, other_surface.pixel_format, other_surface.is_tiled) ==
                std::tie(addr, width, height, stride, pixel_format, is_tiled) &&
-           pixel_format != PixelFormat::Invalid;
+           pixel_format != PixelFormat::Invalid && levels >= other_surface.levels;
 }
 
 bool SurfaceParams::CanSubRect(const SurfaceParams& sub_surface) const {

src/video_core/rasterizer_cache/texture_codec.h

@@ -305,7 +305,9 @@ static constexpr void MortonCopy(u32 width, u32 height, u32 start_offset, u32 en
 
     // If the copy spans multiple tiles, copy the fully aligned tiles in between.
     if (aligned_start_offset < aligned_end_offset) {
-        const u32 buffer_end = tiled_offset + aligned_end_offset - aligned_start_offset;
+        const u32 tile_buffer_size = static_cast<u32>(tiled_buffer.size());
+        const u32 buffer_end =
+            std::min(tiled_offset + aligned_end_offset - aligned_start_offset, tile_buffer_size);
         while (tiled_offset < buffer_end) {
             auto linear_data = linear_buffer.subspan(linear_offset, linear_tile_stride);
             auto tiled_data = tiled_buffer.subspan(tiled_offset, tile_size);