Plexi09/eden-miror
1
0
Fork 0
mirror of https://git.eden-emu.dev/eden-emu/eden synced 2026-05-13 19:38:37 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

[vk] Fast UBO: fix tracking (#2712)

Browse source 
Fixes or mitigates memory errors in TOTK and possibly other games as well.

Credit: Ribbit
Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/2712
Reviewed-by: Lizzie <lizzie@eden-emu.dev>
Reviewed-by: crueter <crueter@eden-emu.dev>
Co-authored-by: MaranBr <maranbr@outlook.com>
Co-committed-by: MaranBr <maranbr@outlook.com>
This commit is contained in:
MaranBr 2025-10-11 06:34:21 +02:00 • committed by crueter
parent 973a65c4c5
commit 91493fa39b
No known key found for this signature in database
GPG key ID: 425ACD2D4830EBC6
2 changed files with 12 additions and 21 deletions
Download patch file Download diff file Expand all files Collapse all files

27
src/video_core/buffer_cache/buffer_cache.h
View file

@ -386,11 +386,9 @@ void BufferCache<P>::BindHostComputeBuffers() {
template <class P> template <class P>
void BufferCache<P>::SetUniformBuffersState(const std::array<u32, NUM_STAGES>& mask, void BufferCache<P>::SetUniformBuffersState(const std::array<u32, NUM_STAGES>& mask,
const UniformBufferSizes* sizes) { const UniformBufferSizes* sizes) {
if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { if (channel_state->enabled_uniform_buffer_masks != mask) {
if (channel_state->enabled_uniform_buffer_masks != mask) { channel_state->fast_bound_uniform_buffers.fill(0);
if constexpr (IS_OPENGL) { if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) {
channel_state->fast_bound_uniform_buffers.fill(0);
}
channel_state->dirty_uniform_buffers.fill(~u32{0}); channel_state->dirty_uniform_buffers.fill(~u32{0});
channel_state->uniform_buffer_binding_sizes.fill({}); channel_state->uniform_buffer_binding_sizes.fill({});
} }
@ -806,7 +804,7 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
channel_state->uniform_buffer_binding_sizes[stage][binding_index] != size; channel_state->uniform_buffer_binding_sizes[stage][binding_index] != size;
if (should_fast_bind) { if (should_fast_bind) {
// We only have to bind when the currently bound buffer is not the fast version // 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; channel_state->uniform_buffer_binding_sizes[stage][binding_index] = size;
runtime.BindFastUniformBuffer(stage, binding_index, size); runtime.BindFastUniformBuffer(stage, binding_index, size);
} }
@ -815,10 +813,8 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
return; return;
} }
} }
if constexpr (IS_OPENGL) { 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;
channel_state->uniform_buffer_binding_sizes[stage][binding_index] = size;
}
// Stream buffer path to avoid stalling on non-Nvidia drivers or Vulkan // Stream buffer path to avoid stalling on non-Nvidia drivers or Vulkan
const std::span<u8> span = runtime.BindMappedUniformBuffer(stage, binding_index, size); const std::span<u8> span = runtime.BindMappedUniformBuffer(stage, binding_index, size);
device_memory.ReadBlockUnsafe(device_addr, span.data(), size); device_memory.ReadBlockUnsafe(device_addr, span.data(), size);
@ -839,9 +835,6 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
} }
const u32 offset = buffer.Offset(device_addr); const u32 offset = buffer.Offset(device_addr);
if constexpr (IS_OPENGL) { 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 // Mark the index as dirty if offset doesn't match
const bool is_copy_bind = offset != 0 && !runtime.SupportsNonZeroUniformOffset(); const bool is_copy_bind = offset != 0 && !runtime.SupportsNonZeroUniformOffset();
channel_state->dirty_uniform_buffers[stage] |= (is_copy_bind ? 1U : 0U) << index; channel_state->dirty_uniform_buffers[stage] |= (is_copy_bind ? 1U : 0U) << index;
@ -855,6 +848,7 @@ void BufferCache<P>::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32
} else { } else {
runtime.BindUniformBuffer(buffer, offset, size); runtime.BindUniformBuffer(buffer, offset, size);
} }
channel_state->fast_bound_uniform_buffers[stage] &= ~(1u << binding_index);
} }
template <class P> template <class P>
@ -1789,12 +1783,7 @@ std::span<u8> BufferCache<P>::ImmediateBuffer(size_t wanted_capacity) {
template <class P> template <class P>
bool BufferCache<P>::HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept { bool BufferCache<P>::HasFastUniformBufferBound(size_t stage, u32 binding_index) const noexcept {
if constexpr (IS_OPENGL) { return ((channel_state->fast_bound_uniform_buffers[stage] >> binding_index) & 1u) != 0;
return ((channel_state->fast_bound_uniform_buffers[stage] >> binding_index) & 1) != 0;
} else {
// Only OpenGL has fast uniform buffers
return false;
}
} }
template <class P> template <class P>

6
src/video_core/buffer_cache/buffer_cache_base.h
View file

@ -54,6 +54,8 @@ constexpr u32 NUM_STORAGE_BUFFERS = 16;
constexpr u32 NUM_TEXTURE_BUFFERS = 32; constexpr u32 NUM_TEXTURE_BUFFERS = 32;
constexpr u32 NUM_STAGES = 5; constexpr u32 NUM_STAGES = 5;
static_assert(NUM_GRAPHICS_UNIFORM_BUFFERS <= 32, "fast bitmask must fit u32");
using UniformBufferSizes = std::array<std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES>; using UniformBufferSizes = std::array<std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFERS>, NUM_STAGES>;
using ComputeUniformBufferSizes = std::array<u32, NUM_COMPUTE_UNIFORM_BUFFERS>; using ComputeUniformBufferSizes = std::array<u32, NUM_COMPUTE_UNIFORM_BUFFERS>;
@ -137,8 +139,8 @@ public:
u32 written_compute_texture_buffers = 0; u32 written_compute_texture_buffers = 0;
u32 image_compute_texture_buffers = 0; u32 image_compute_texture_buffers = 0;
std::array<u32, 16> uniform_cache_hits{}; std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFERS> uniform_cache_hits{};
std::array<u32, 16> uniform_cache_shots{}; std::array<u32, NUM_GRAPHICS_UNIFORM_BUFFERS> uniform_cache_shots{};
u32 uniform_buffer_skip_cache_size = DEFAULT_SKIP_CACHE_SIZE; u32 uniform_buffer_skip_cache_size = DEFAULT_SKIP_CACHE_SIZE;