video_core: cheaper small GPU reads, buffer usage tracking and compute uniforms

- MemoryManager::ReadBlockUnsafe reads within one 4 KiB page through a
  single translation (as Read<T> does) instead of the page walk:
  texture and sampler descriptors, several per draw (2.6% -> 0.1% of
  title 1's GPU thread).
- UsageTracker skips ranges it tracked just before (until Reset): draws
  bind the same multi-megabyte storage buffers again and again, and
  each Track wrote a word per 4 KiB (2.0% -> 0.7%).
- Small compute uniform buffers the GPU has not written are streamed,
  as graphics ones are; synchronizing them re-protected pages an inline
  upload had just unprotected, two VirtualProtect calls per dispatch
  (1.9% -> 0.7%, page protection 3.2% -> 2.3%).

1 files changed, 8 insertions, 1 deletions

diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h
index 4df855c0fe..ad2e200438 100644
--- a/src/video_core/buffer_cache/buffer_cache.h
+++ b/src/video_core/buffer_cache/buffer_cache.h
@@ -1211,8 +1211,15 @@ void BufferCache<P>::BindHostComputeUniformBuffers() {
                 return alignment > 1 && (offset % alignment) != 0;
             }
         }();
+        // Small buffers the GPU has not written are streamed, as graphics ones are: synchronizing
+        // them made every dispatch re-protect pages an inline upload had just unprotected (two
+        // VirtualProtect calls per dispatch, ~4,000 dispatches a second in title 1).
+        const bool use_fast_buffer =
+            needs_alignment_stream ||
+            (has_host_buffer && size <= channel_state->uniform_buffer_skip_cache_size &&
+             !memory_tracker.IsRegionGpuModified(binding.device_addr, size));
         if constexpr (!IS_OPENGL) {
-            if (needs_alignment_stream) {
+            if (use_fast_buffer) {
                 const std::span<u8> span =
                     runtime.BindMappedUniformBuffer(0, binding_index, size);
                 device_memory.ReadBlockUnsafe(binding.device_addr, span.data(), size);