From 8078990b9b95c5bb2c432abc9012c2738f8a451a Mon Sep 17 00:00:00 2001 From: Ribbit Date: Wed, 8 Oct 2025 05:39:08 +0200 Subject: [vk] Fast UBO: fix tracking, resize heuristics, add debug guard (#2695) Co-authored-by: Ribbit Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/2695 Reviewed-by: CamilleLaVey Co-authored-by: Ribbit Co-committed-by: Ribbit --- src/video_core/buffer_cache/buffer_cache.h | 42 ++++++++++++++++-------------- 1 file changed, 22 insertions(+), 20 deletions(-) (limited to 'src/video_core/buffer_cache/buffer_cache.h') diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 5223afe937..388c8034c5 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -386,11 +386,10 @@ void BufferCache

::BindHostComputeBuffers() { template void BufferCache

::SetUniformBuffersState(const std::array& mask, const UniformBufferSizes* sizes) { - if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { - if (channel_state->enabled_uniform_buffer_masks != mask) { - if constexpr (IS_OPENGL) { - channel_state->fast_bound_uniform_buffers.fill(0); - } + const bool mask_changed = channel_state->enabled_uniform_buffer_masks != mask; + if (mask_changed) { + channel_state->fast_bound_uniform_buffers.fill(0); + if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { channel_state->dirty_uniform_buffers.fill(~u32{0}); channel_state->uniform_buffer_binding_sizes.fill({}); } @@ -806,7 +805,7 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 channel_state->uniform_buffer_binding_sizes[stage][binding_index] != size; if (should_fast_bind) { // We only have to bind when the currently bound buffer is not the fast version - channel_state->fast_bound_uniform_buffers[stage] |= 1U << binding_index; + channel_state->fast_bound_uniform_buffers[stage] |= 1u << binding_index; channel_state->uniform_buffer_binding_sizes[stage][binding_index] = size; runtime.BindFastUniformBuffer(stage, binding_index, size); } @@ -815,13 +814,22 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 return; } } - if constexpr (IS_OPENGL) { - channel_state->fast_bound_uniform_buffers[stage] |= 1U << binding_index; - channel_state->uniform_buffer_binding_sizes[stage][binding_index] = size; - } + channel_state->fast_bound_uniform_buffers[stage] |= 1u << binding_index; + channel_state->uniform_buffer_binding_sizes[stage][binding_index] = size; // Stream buffer path to avoid stalling on non-Nvidia drivers or Vulkan const std::span span = runtime.BindMappedUniformBuffer(stage, binding_index, size); +#ifdef YUZU_DEBUG + ASSERT(binding_index < NUM_GRAPHICS_UNIFORM_BUFFERS); + ASSERT(span.size() >= size && "UBO stream span too small"); + if (!device_memory.ReadBlockFastChecked(device_addr, span.data(), size)) { + LOG_CRITICAL(Render, "DeviceMemory OOB/unmapped: addr=0x{:x} size={}", device_addr, size); + channel_state->fast_bound_uniform_buffers[stage] &= ~(1u << binding_index); + ASSERT(false); + return; + } +#else device_memory.ReadBlockUnsafe(device_addr, span.data(), size); +#endif return; } // Classic cached path @@ -830,7 +838,8 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 } // Skip binding if it's not needed and if the bound buffer is not the fast version // This exists to avoid instances where the fast buffer is bound and a GPU write happens - needs_bind |= HasFastUniformBufferBound(stage, binding_index); + const bool was_fast_bound = HasFastUniformBufferBound(stage, binding_index); + needs_bind |= was_fast_bound; if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { needs_bind |= channel_state->uniform_buffer_binding_sizes[stage][binding_index] != size; } @@ -839,9 +848,6 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 } const u32 offset = buffer.Offset(device_addr); if constexpr (IS_OPENGL) { - // Fast buffer will be unbound - channel_state->fast_bound_uniform_buffers[stage] &= ~(1U << binding_index); - // Mark the index as dirty if offset doesn't match const bool is_copy_bind = offset != 0 && !runtime.SupportsNonZeroUniformOffset(); channel_state->dirty_uniform_buffers[stage] |= (is_copy_bind ? 1U : 0U) << index; @@ -855,6 +861,7 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 } else { runtime.BindUniformBuffer(buffer, offset, size); } + channel_state->fast_bound_uniform_buffers[stage] &= ~(1u << binding_index); } template @@ -1789,12 +1796,7 @@ std::span BufferCache

::ImmediateBuffer(size_t wanted_capacity) { template bool BufferCache

::HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept { - if constexpr (IS_OPENGL) { - return ((channel_state->fast_bound_uniform_buffers[stage] >> binding_index) & 1) != 0; - } else { - // Only OpenGL has fast uniform buffers - return false; - } + return ((channel_state->fast_bound_uniform_buffers[stage] >> binding_index) & 1u) != 0; } template -- cgit v1.2.3