nx2: memo for compute macros; one lock per inline upload; cheaper shader invalidation

- Compute engine macros replay repeated calls like 3D ones (title 1's
  compute macro path fell from ~27% to ~14% of the GPU thread).
- AccelerateInlineToMemory takes the buffer and texture cache mutexes
  together instead of one after the other.
- ShaderCache::InvalidatePagesInRegion looks at the rejected programs
  only when the write can reach one (bounds kept as they are added);
  it ran a locked scan for every inline upload and pre-draw
  invalidation.

Inline uploads went from ~10% to ~4% of the GPU thread.

1 files changed, 4 insertions, 0 deletions

diff --git a/src/video_core/shader_cache.h b/src/video_core/shader_cache.h
index 9ffb7bb44d..f4179dd435 100644
--- a/src/video_core/shader_cache.h
+++ b/src/video_core/shader_cache.h
@@ -7,6 +7,7 @@
 #pragma once
 
 #include <array>
+#include <atomic>
 #include <exception>
 #include <memory>
 #include <mutex>
@@ -165,6 +166,9 @@ private:
     /// first error instead of lowering again; dropped when guest memory there is written.
     std::mutex rejected_mutex;
     ::Common::unordered_map<u64, std::exception_ptr> rejected_native;
+    /// Bounds of every address ever rejected (never shrinks): a quick test for invalidations.
+    std::atomic<VAddr> rejected_lowest{~VAddr{}};
+    std::atomic<VAddr> rejected_highest{};
 };
 
 } // namespace VideoCommon