video_core: less waiting on the GPU caches' locks

- The texture and buffer caches' mutexes spin briefly before sleeping
  (Common::SpinRecursiveMutex): the GPU thread, the fencing thread and
  CPU write traps take them tens of thousands of times a second for
  microseconds, and a sleeping waiter's wake-up took longer than that.
  Title 1's GPU thread slept 7.7% of its time there, now 3.7%.
- Fences commit and pop texture and buffer downloads separately, and
  only when there are any (one fence per compute dispatch in Switch 2
  titles): the fencing thread no longer holds both caches while it
  copies buffer downloads.
- Inline uploads and pre-draw invalidation take the caches one at a
  time again: holding one while std::lock waited for the other waited
  longer.

1 files changed, 2 insertions, 1 deletions

diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h
index e4aa5896e5..57011752cc 100644
--- a/src/video_core/buffer_cache/buffer_cache_base.h
+++ b/src/video_core/buffer_cache/buffer_cache_base.h
@@ -21,6 +21,7 @@
 #include <boost/container/small_vector.hpp>
 
 #include "common/common_types.h"
+#include "common/spin_recursive_mutex.h"
 #include "common/div_ceil.h"
 #include "common/literals.h"
 #include "common/lru_cache.h"
@@ -339,7 +340,7 @@ public:
         } while (channel_state->has_deleted_buffers);
     }
 
-    std::recursive_mutex mutex;
+    Common::SpinRecursiveMutex mutex;
     Runtime& runtime;
     bool any_buffer_uploaded = false;