diff options
Diffstat (limited to 'src/video_core/buffer_cache')
| -rw-r--r-- | src/video_core/buffer_cache/buffer_cache.h | 9 | ||||
| -rw-r--r-- | src/video_core/buffer_cache/usage_tracker.h | 17 |
2 files changed, 25 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); diff --git a/src/video_core/buffer_cache/usage_tracker.h b/src/video_core/buffer_cache/usage_tracker.h index 6f84e463e3..3e2da8bc0b 100644 --- a/src/video_core/buffer_cache/usage_tracker.h +++ b/src/video_core/buffer_cache/usage_tracker.h @@ -6,6 +6,10 @@ #pragma once +#include <array> +#include <utility> +#include <vector> + #include "common/alignment.h" #include "common/common_types.h" @@ -24,9 +28,19 @@ public: void Reset() noexcept { std::ranges::fill(pages, 0ULL); + recent.fill({~u64{0}, 0}); } void Track(u64 offset, u64 size) noexcept { + // Tracking is idempotent until Reset: skip ranges tracked just before. Draws bind the + // same (often multi-megabyte) storage buffers again and again, and each Track wrote a + // word per 4 KiB of the range. + for (const auto& [recent_offset, recent_size] : recent) { + if (recent_offset == offset && recent_size == size) { + return; + } + } + recent[recent_cursor++ % recent.size()] = {offset, size}; const size_t page = offset >> BUFFER_PAGE_SHIFT; const size_t page_end = (offset + size) >> BUFFER_PAGE_SHIFT; if (page_end < page || page_end >= pages.size()) { @@ -83,6 +97,9 @@ private: private: std::vector<u64> pages; + /// Ranges tracked last, (offset, size). + std::array<std::pair<u64, u64>, 4> recent{{{~u64{0}, 0}, {~u64{0}, 0}, {~u64{0}, 0}, {~u64{0}, 0}}}; + size_t recent_cursor{}; }; } // namespace VideoCommon |