GPU/HW: Don't use shader for self-overlapping copies

If we use the shader here, we're going to immediately end the
render pass since the area we read and wrote to is going to be
dirty again.

This is problematic on Mali where the render passes are really
expensive. Test case: Dino Crisis 2 - open menu

src/core/gpu_hw.cpp

@@ -3463,11 +3463,12 @@ void GPU_HW::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32
   const GSVector4i dst_bounds = GetVRAMTransferBounds(dst_x, dst_y, width, height);
   const bool intersect_with_draw = m_vram_dirty_draw_rect.rintersects(src_bounds);
   const bool intersect_with_write = m_vram_dirty_write_rect.rintersects(src_bounds);
+  const bool overlaps_with_self = src_bounds.rintersects(dst_bounds);
   const bool use_shader =
     (set_mask || check_mask || ((src_x % VRAM_WIDTH) + width) > VRAM_WIDTH ||
      ((src_y % VRAM_HEIGHT) + height) > VRAM_HEIGHT || ((dst_x % VRAM_WIDTH) + width) > VRAM_WIDTH ||
      ((dst_y % VRAM_HEIGHT) + height) > VRAM_HEIGHT) ||
-    (!intersect_with_draw && !intersect_with_write);
+    (!intersect_with_draw && !intersect_with_write && !overlaps_with_self);
 
   // If we're copying a region that hasn't been drawn to, and we're using the TC, we can do it in local memory.
   if (m_use_texture_cache && !GPUTextureCache::IsRectDrawn(src_bounds))
@@ -3535,7 +3536,6 @@ void GPU_HW::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32
   }
 
   GPUTexture* src_tex = m_vram_texture.get();
-  const bool overlaps_with_self = src_bounds.rintersects(dst_bounds);
   if (!g_gpu_device->GetFeatures().texture_copy_to_self || overlaps_with_self)
   {
     src_tex = m_vram_read_texture.get();