From b0bea13ed8422119cc4f09763be095cbc3762795 Mon Sep 17 00:00:00 2001 From: Kelebek1 Date: Sat, 27 May 2023 15:45:36 +0100 Subject: [PATCH] Move buffer bindings to per-channel state --- src/video_core/buffer_cache/buffer_cache.cpp | 4 + src/video_core/buffer_cache/buffer_cache.h | 291 +++++++++--------- .../buffer_cache/buffer_cache_base.h | 141 ++++----- .../renderer_opengl/gl_buffer_cache.cpp | 2 +- 4 files changed, 228 insertions(+), 210 deletions(-) diff --git a/src/video_core/buffer_cache/buffer_cache.cpp b/src/video_core/buffer_cache/buffer_cache.cpp index 40db243d2..4b4f7061b 100644 --- a/src/video_core/buffer_cache/buffer_cache.cpp +++ b/src/video_core/buffer_cache/buffer_cache.cpp @@ -2,6 +2,8 @@ // SPDX-License-Identifier: GPL-3.0-or-later #include "common/microprofile.h" +#include "video_core/buffer_cache/buffer_cache_base.h" +#include "video_core/control/channel_state_cache.inc" namespace VideoCommon { @@ -9,4 +11,6 @@ MICROPROFILE_DEFINE(GPU_PrepareBuffers, "GPU", "Prepare buffers", MP_RGB(224, 12 MICROPROFILE_DEFINE(GPU_BindUploadBuffers, "GPU", "Bind and upload buffers", MP_RGB(224, 128, 128)); MICROPROFILE_DEFINE(GPU_DownloadMemory, "GPU", "Download buffers", MP_RGB(224, 128, 128)); +template class VideoCommon::ChannelSetupCaches; + } // namespace VideoCommon diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 65494097b..c336be707 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -64,17 +64,22 @@ void BufferCache

::RunGarbageCollector() { template void BufferCache

::TickFrame() { // Calculate hits and shots and move hit bits to the right - const u32 hits = std::reduce(uniform_cache_hits.begin(), uniform_cache_hits.end()); - const u32 shots = std::reduce(uniform_cache_shots.begin(), uniform_cache_shots.end()); - std::copy_n(uniform_cache_hits.begin(), uniform_cache_hits.size() - 1, - uniform_cache_hits.begin() + 1); - std::copy_n(uniform_cache_shots.begin(), uniform_cache_shots.size() - 1, - uniform_cache_shots.begin() + 1); - uniform_cache_hits[0] = 0; - uniform_cache_shots[0] = 0; + + const u32 hits = std::reduce(channel_state->uniform_cache_hits.begin(), + channel_state->uniform_cache_hits.end()); + const u32 shots = std::reduce(channel_state->uniform_cache_shots.begin(), + channel_state->uniform_cache_shots.end()); + std::copy_n(channel_state->uniform_cache_hits.begin(), + channel_state->uniform_cache_hits.size() - 1, + channel_state->uniform_cache_hits.begin() + 1); + std::copy_n(channel_state->uniform_cache_shots.begin(), + channel_state->uniform_cache_shots.size() - 1, + channel_state->uniform_cache_shots.begin() + 1); + channel_state->uniform_cache_hits[0] = 0; + channel_state->uniform_cache_shots[0] = 0; const bool skip_preferred = hits * 256 < shots * 251; - uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0; + channel_state->uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0; // If we can obtain the memory info, use it instead of the estimate. if (runtime.CanReportMemoryUsage()) { @@ -164,10 +169,10 @@ bool BufferCache

::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am BufferId buffer_a; BufferId buffer_b; do { - has_deleted_buffers = false; + channel_state->has_deleted_buffers = false; buffer_a = FindBuffer(*cpu_src_address, static_cast(amount)); buffer_b = FindBuffer(*cpu_dest_address, static_cast(amount)); - } while (has_deleted_buffers); + } while (channel_state->has_deleted_buffers); auto& src_buffer = slot_buffers[buffer_a]; auto& dest_buffer = slot_buffers[buffer_b]; SynchronizeBuffer(src_buffer, *cpu_src_address, static_cast(amount)); @@ -272,30 +277,30 @@ void BufferCache

::BindGraphicsUniformBuffer(size_t stage, u32 index, GPUVAddr .size = size, .buffer_id = BufferId{}, }; - uniform_buffers[stage][index] = binding; + channel_state->uniform_buffers[stage][index] = binding; } template void BufferCache

::DisableGraphicsUniformBuffer(size_t stage, u32 index) { - uniform_buffers[stage][index] = NULL_BINDING; + channel_state->uniform_buffers[stage][index] = NULL_BINDING; } template void BufferCache

::UpdateGraphicsBuffers(bool is_indexed) { MICROPROFILE_SCOPE(GPU_PrepareBuffers); do { - has_deleted_buffers = false; + channel_state->has_deleted_buffers = false; DoUpdateGraphicsBuffers(is_indexed); - } while (has_deleted_buffers); + } while (channel_state->has_deleted_buffers); } template void BufferCache

::UpdateComputeBuffers() { MICROPROFILE_SCOPE(GPU_PrepareBuffers); do { - has_deleted_buffers = false; + channel_state->has_deleted_buffers = false; DoUpdateComputeBuffers(); - } while (has_deleted_buffers); + } while (channel_state->has_deleted_buffers); } template @@ -338,98 +343,102 @@ template void BufferCache

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

::SetComputeUniformBufferState(u32 mask, const ComputeUniformBufferSizes* sizes) { - enabled_compute_uniform_buffer_mask = mask; - compute_uniform_buffer_sizes = sizes; + channel_state->enabled_compute_uniform_buffer_mask = mask; + channel_state->compute_uniform_buffer_sizes = sizes; } template void BufferCache

::UnbindGraphicsStorageBuffers(size_t stage) { - enabled_storage_buffers[stage] = 0; - written_storage_buffers[stage] = 0; + channel_state->enabled_storage_buffers[stage] = 0; + channel_state->written_storage_buffers[stage] = 0; } template void BufferCache

::BindGraphicsStorageBuffer(size_t stage, size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset, bool is_written) { - enabled_storage_buffers[stage] |= 1U << ssbo_index; - written_storage_buffers[stage] |= (is_written ? 1U : 0U) << ssbo_index; + channel_state->enabled_storage_buffers[stage] |= 1U << ssbo_index; + channel_state->written_storage_buffers[stage] |= (is_written ? 1U : 0U) << ssbo_index; const auto& cbufs = maxwell3d->state.shader_stages[stage]; const GPUVAddr ssbo_addr = cbufs.const_buffers[cbuf_index].address + cbuf_offset; - storage_buffers[stage][ssbo_index] = StorageBufferBinding(ssbo_addr, cbuf_index, is_written); + channel_state->storage_buffers[stage][ssbo_index] = + StorageBufferBinding(ssbo_addr, cbuf_index, is_written); } template void BufferCache

::UnbindGraphicsTextureBuffers(size_t stage) { - enabled_texture_buffers[stage] = 0; - written_texture_buffers[stage] = 0; - image_texture_buffers[stage] = 0; + channel_state->enabled_texture_buffers[stage] = 0; + channel_state->written_texture_buffers[stage] = 0; + channel_state->image_texture_buffers[stage] = 0; } template void BufferCache

::BindGraphicsTextureBuffer(size_t stage, size_t tbo_index, GPUVAddr gpu_addr, u32 size, PixelFormat format, bool is_written, bool is_image) { - enabled_texture_buffers[stage] |= 1U << tbo_index; - written_texture_buffers[stage] |= (is_written ? 1U : 0U) << tbo_index; + channel_state->enabled_texture_buffers[stage] |= 1U << tbo_index; + channel_state->written_texture_buffers[stage] |= (is_written ? 1U : 0U) << tbo_index; if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) { - image_texture_buffers[stage] |= (is_image ? 1U : 0U) << tbo_index; + channel_state->image_texture_buffers[stage] |= (is_image ? 1U : 0U) << tbo_index; } - texture_buffers[stage][tbo_index] = GetTextureBufferBinding(gpu_addr, size, format); + channel_state->texture_buffers[stage][tbo_index] = + GetTextureBufferBinding(gpu_addr, size, format); } template void BufferCache

::UnbindComputeStorageBuffers() { - enabled_compute_storage_buffers = 0; - written_compute_storage_buffers = 0; - image_compute_texture_buffers = 0; + channel_state->enabled_compute_storage_buffers = 0; + channel_state->written_compute_storage_buffers = 0; + channel_state->image_compute_texture_buffers = 0; } template void BufferCache

::BindComputeStorageBuffer(size_t ssbo_index, u32 cbuf_index, u32 cbuf_offset, bool is_written) { - enabled_compute_storage_buffers |= 1U << ssbo_index; - written_compute_storage_buffers |= (is_written ? 1U : 0U) << ssbo_index; + channel_state->enabled_compute_storage_buffers |= 1U << ssbo_index; + channel_state->written_compute_storage_buffers |= (is_written ? 1U : 0U) << ssbo_index; const auto& launch_desc = kepler_compute->launch_description; ASSERT(((launch_desc.const_buffer_enable_mask >> cbuf_index) & 1) != 0); const auto& cbufs = launch_desc.const_buffer_config; const GPUVAddr ssbo_addr = cbufs[cbuf_index].Address() + cbuf_offset; - compute_storage_buffers[ssbo_index] = StorageBufferBinding(ssbo_addr, cbuf_index, is_written); + channel_state->compute_storage_buffers[ssbo_index] = + StorageBufferBinding(ssbo_addr, cbuf_index, is_written); } template void BufferCache

::UnbindComputeTextureBuffers() { - enabled_compute_texture_buffers = 0; - written_compute_texture_buffers = 0; - image_compute_texture_buffers = 0; + channel_state->enabled_compute_texture_buffers = 0; + channel_state->written_compute_texture_buffers = 0; + channel_state->image_compute_texture_buffers = 0; } template void BufferCache

::BindComputeTextureBuffer(size_t tbo_index, GPUVAddr gpu_addr, u32 size, PixelFormat format, bool is_written, bool is_image) { - enabled_compute_texture_buffers |= 1U << tbo_index; - written_compute_texture_buffers |= (is_written ? 1U : 0U) << tbo_index; + channel_state->enabled_compute_texture_buffers |= 1U << tbo_index; + channel_state->written_compute_texture_buffers |= (is_written ? 1U : 0U) << tbo_index; if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) { - image_compute_texture_buffers |= (is_image ? 1U : 0U) << tbo_index; + channel_state->image_compute_texture_buffers |= (is_image ? 1U : 0U) << tbo_index; } - compute_texture_buffers[tbo_index] = GetTextureBufferBinding(gpu_addr, size, format); + channel_state->compute_texture_buffers[tbo_index] = + GetTextureBufferBinding(gpu_addr, size, format); } template @@ -672,10 +681,10 @@ bool BufferCache

::IsRegionCpuModified(VAddr addr, size_t size) { template void BufferCache

::BindHostIndexBuffer() { - Buffer& buffer = slot_buffers[index_buffer.buffer_id]; - TouchBuffer(buffer, index_buffer.buffer_id); - const u32 offset = buffer.Offset(index_buffer.cpu_addr); - const u32 size = index_buffer.size; + Buffer& buffer = slot_buffers[channel_state->index_buffer.buffer_id]; + TouchBuffer(buffer, channel_state->index_buffer.buffer_id); + const u32 offset = buffer.Offset(channel_state->index_buffer.cpu_addr); + const u32 size = channel_state->index_buffer.size; const auto& draw_state = maxwell3d->draw_manager->GetDrawState(); if (!draw_state.inline_index_draw_indexes.empty()) [[unlikely]] { if constexpr (USE_MEMORY_MAPS) { @@ -689,7 +698,7 @@ void BufferCache

::BindHostIndexBuffer() { buffer.ImmediateUpload(0, draw_state.inline_index_draw_indexes); } } else { - SynchronizeBuffer(buffer, index_buffer.cpu_addr, size); + SynchronizeBuffer(buffer, channel_state->index_buffer.cpu_addr, size); } if constexpr (HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) { const u32 new_offset = @@ -706,7 +715,7 @@ template void BufferCache

::BindHostVertexBuffers() { auto& flags = maxwell3d->dirty.flags; for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) { - const Binding& binding = vertex_buffers[index]; + const Binding& binding = channel_state->vertex_buffers[index]; Buffer& buffer = slot_buffers[binding.buffer_id]; TouchBuffer(buffer, binding.buffer_id); SynchronizeBuffer(buffer, binding.cpu_addr, binding.size); @@ -729,19 +738,19 @@ void BufferCache

::BindHostDrawIndirectBuffers() { SynchronizeBuffer(buffer, binding.cpu_addr, binding.size); }; if (current_draw_indirect->include_count) { - bind_buffer(count_buffer_binding); + bind_buffer(channel_state->count_buffer_binding); } - bind_buffer(indirect_buffer_binding); + bind_buffer(channel_state->indirect_buffer_binding); } template void BufferCache

::BindHostGraphicsUniformBuffers(size_t stage) { u32 dirty = ~0U; if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { - dirty = std::exchange(dirty_uniform_buffers[stage], 0); + dirty = std::exchange(channel_state->dirty_uniform_buffers[stage], 0); } u32 binding_index = 0; - ForEachEnabledBit(enabled_uniform_buffer_masks[stage], [&](u32 index) { + ForEachEnabledBit(channel_state->enabled_uniform_buffer_masks[stage], [&](u32 index) { const bool needs_bind = ((dirty >> index) & 1) != 0; BindHostGraphicsUniformBuffer(stage, index, binding_index, needs_bind); if constexpr (NEEDS_BIND_UNIFORM_INDEX) { @@ -753,13 +762,13 @@ void BufferCache

::BindHostGraphicsUniformBuffers(size_t stage) { template void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 binding_index, bool needs_bind) { - const Binding& binding = uniform_buffers[stage][index]; + const Binding& binding = channel_state->uniform_buffers[stage][index]; const VAddr cpu_addr = binding.cpu_addr; - const u32 size = std::min(binding.size, (*uniform_buffer_sizes)[stage][index]); + const u32 size = std::min(binding.size, (*channel_state->uniform_buffer_sizes)[stage][index]); Buffer& buffer = slot_buffers[binding.buffer_id]; TouchBuffer(buffer, binding.buffer_id); const bool use_fast_buffer = binding.buffer_id != NULL_BUFFER_ID && - size <= uniform_buffer_skip_cache_size && + size <= channel_state->uniform_buffer_skip_cache_size && !memory_tracker.IsRegionGpuModified(cpu_addr, size); if (use_fast_buffer) { if constexpr (IS_OPENGL) { @@ -767,11 +776,11 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 // Fast path for Nvidia const bool should_fast_bind = !HasFastUniformBufferBound(stage, binding_index) || - uniform_buffer_binding_sizes[stage][binding_index] != size; + 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 - fast_bound_uniform_buffers[stage] |= 1U << binding_index; - 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; runtime.BindFastUniformBuffer(stage, binding_index, size); } const auto span = ImmediateBufferWithData(cpu_addr, size); @@ -780,8 +789,8 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 } } if constexpr (IS_OPENGL) { - fast_bound_uniform_buffers[stage] |= 1U << binding_index; - 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); @@ -791,15 +800,15 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 // Classic cached path const bool sync_cached = SynchronizeBuffer(buffer, cpu_addr, size); if (sync_cached) { - ++uniform_cache_hits[0]; + ++channel_state->uniform_cache_hits[0]; } - ++uniform_cache_shots[0]; + ++channel_state->uniform_cache_shots[0]; // 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); if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { - needs_bind |= uniform_buffer_binding_sizes[stage][binding_index] != size; + needs_bind |= channel_state->uniform_buffer_binding_sizes[stage][binding_index] != size; } if (!needs_bind) { return; @@ -807,14 +816,14 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 const u32 offset = buffer.Offset(cpu_addr); if constexpr (IS_OPENGL) { // Fast buffer will be unbound - fast_bound_uniform_buffers[stage] &= ~(1U << binding_index); + 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(); - dirty_uniform_buffers[stage] |= (is_copy_bind ? 1U : 0U) << index; + channel_state->dirty_uniform_buffers[stage] |= (is_copy_bind ? 1U : 0U) << index; } if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { - uniform_buffer_binding_sizes[stage][binding_index] = size; + channel_state->uniform_buffer_binding_sizes[stage][binding_index] = size; } if constexpr (NEEDS_BIND_UNIFORM_INDEX) { runtime.BindUniformBuffer(stage, binding_index, buffer, offset, size); @@ -826,15 +835,15 @@ void BufferCache

::BindHostGraphicsUniformBuffer(size_t stage, u32 index, u32 template void BufferCache

::BindHostGraphicsStorageBuffers(size_t stage) { u32 binding_index = 0; - ForEachEnabledBit(enabled_storage_buffers[stage], [&](u32 index) { - const Binding& binding = storage_buffers[stage][index]; + ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) { + const Binding& binding = channel_state->storage_buffers[stage][index]; Buffer& buffer = slot_buffers[binding.buffer_id]; TouchBuffer(buffer, binding.buffer_id); const u32 size = binding.size; SynchronizeBuffer(buffer, binding.cpu_addr, size); const u32 offset = buffer.Offset(binding.cpu_addr); - const bool is_written = ((written_storage_buffers[stage] >> index) & 1) != 0; + const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0; if constexpr (NEEDS_BIND_STORAGE_INDEX) { runtime.BindStorageBuffer(stage, binding_index, buffer, offset, size, is_written); ++binding_index; @@ -846,8 +855,8 @@ void BufferCache

::BindHostGraphicsStorageBuffers(size_t stage) { template void BufferCache

::BindHostGraphicsTextureBuffers(size_t stage) { - ForEachEnabledBit(enabled_texture_buffers[stage], [&](u32 index) { - const TextureBufferBinding& binding = texture_buffers[stage][index]; + ForEachEnabledBit(channel_state->enabled_texture_buffers[stage], [&](u32 index) { + const TextureBufferBinding& binding = channel_state->texture_buffers[stage][index]; Buffer& buffer = slot_buffers[binding.buffer_id]; const u32 size = binding.size; SynchronizeBuffer(buffer, binding.cpu_addr, size); @@ -855,7 +864,7 @@ void BufferCache

::BindHostGraphicsTextureBuffers(size_t stage) { const u32 offset = buffer.Offset(binding.cpu_addr); const PixelFormat format = binding.format; if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) { - if (((image_texture_buffers[stage] >> index) & 1) != 0) { + if (((channel_state->image_texture_buffers[stage] >> index) & 1) != 0) { runtime.BindImageBuffer(buffer, offset, size, format); } else { runtime.BindTextureBuffer(buffer, offset, size, format); @@ -872,7 +881,7 @@ void BufferCache

::BindHostTransformFeedbackBuffers() { return; } for (u32 index = 0; index < NUM_TRANSFORM_FEEDBACK_BUFFERS; ++index) { - const Binding& binding = transform_feedback_buffers[index]; + const Binding& binding = channel_state->transform_feedback_buffers[index]; Buffer& buffer = slot_buffers[binding.buffer_id]; TouchBuffer(buffer, binding.buffer_id); const u32 size = binding.size; @@ -887,15 +896,16 @@ template void BufferCache

::BindHostComputeUniformBuffers() { if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { // Mark all uniform buffers as dirty - dirty_uniform_buffers.fill(~u32{0}); - fast_bound_uniform_buffers.fill(0); + channel_state->dirty_uniform_buffers.fill(~u32{0}); + channel_state->fast_bound_uniform_buffers.fill(0); } u32 binding_index = 0; - ForEachEnabledBit(enabled_compute_uniform_buffer_mask, [&](u32 index) { - const Binding& binding = compute_uniform_buffers[index]; + ForEachEnabledBit(channel_state->enabled_compute_uniform_buffer_mask, [&](u32 index) { + const Binding& binding = channel_state->compute_uniform_buffers[index]; Buffer& buffer = slot_buffers[binding.buffer_id]; TouchBuffer(buffer, binding.buffer_id); - const u32 size = std::min(binding.size, (*compute_uniform_buffer_sizes)[index]); + const u32 size = + std::min(binding.size, (*channel_state->compute_uniform_buffer_sizes)[index]); SynchronizeBuffer(buffer, binding.cpu_addr, size); const u32 offset = buffer.Offset(binding.cpu_addr); @@ -911,15 +921,16 @@ void BufferCache

::BindHostComputeUniformBuffers() { template void BufferCache

::BindHostComputeStorageBuffers() { u32 binding_index = 0; - ForEachEnabledBit(enabled_compute_storage_buffers, [&](u32 index) { - const Binding& binding = compute_storage_buffers[index]; + ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) { + const Binding& binding = channel_state->compute_storage_buffers[index]; Buffer& buffer = slot_buffers[binding.buffer_id]; TouchBuffer(buffer, binding.buffer_id); const u32 size = binding.size; SynchronizeBuffer(buffer, binding.cpu_addr, size); const u32 offset = buffer.Offset(binding.cpu_addr); - const bool is_written = ((written_compute_storage_buffers >> index) & 1) != 0; + const bool is_written = + ((channel_state->written_compute_storage_buffers >> index) & 1) != 0; if constexpr (NEEDS_BIND_STORAGE_INDEX) { runtime.BindComputeStorageBuffer(binding_index, buffer, offset, size, is_written); ++binding_index; @@ -931,8 +942,8 @@ void BufferCache

::BindHostComputeStorageBuffers() { template void BufferCache

::BindHostComputeTextureBuffers() { - ForEachEnabledBit(enabled_compute_texture_buffers, [&](u32 index) { - const TextureBufferBinding& binding = compute_texture_buffers[index]; + ForEachEnabledBit(channel_state->enabled_compute_texture_buffers, [&](u32 index) { + const TextureBufferBinding& binding = channel_state->compute_texture_buffers[index]; Buffer& buffer = slot_buffers[binding.buffer_id]; const u32 size = binding.size; SynchronizeBuffer(buffer, binding.cpu_addr, size); @@ -940,7 +951,7 @@ void BufferCache

::BindHostComputeTextureBuffers() { const u32 offset = buffer.Offset(binding.cpu_addr); const PixelFormat format = binding.format; if constexpr (SEPARATE_IMAGE_BUFFERS_BINDINGS) { - if (((image_compute_texture_buffers >> index) & 1) != 0) { + if (((channel_state->image_compute_texture_buffers >> index) & 1) != 0) { runtime.BindImageBuffer(buffer, offset, size, format); } else { runtime.BindTextureBuffer(buffer, offset, size, format); @@ -954,7 +965,7 @@ void BufferCache

::BindHostComputeTextureBuffers() { template void BufferCache

::DoUpdateGraphicsBuffers(bool is_indexed) { do { - has_deleted_buffers = false; + channel_state->has_deleted_buffers = false; if (is_indexed) { UpdateIndexBuffer(); } @@ -968,7 +979,7 @@ void BufferCache

::DoUpdateGraphicsBuffers(bool is_indexed) { if (current_draw_indirect) { UpdateDrawIndirect(); } - } while (has_deleted_buffers); + } while (channel_state->has_deleted_buffers); } template @@ -999,7 +1010,7 @@ void BufferCache

::UpdateIndexBuffer() { slot_buffers.erase(inline_buffer_id); inline_buffer_id = CreateBuffer(0, buffer_size); } - index_buffer = Binding{ + channel_state->index_buffer = Binding{ .cpu_addr = 0, .size = inline_index_size, .buffer_id = inline_buffer_id, @@ -1015,10 +1026,10 @@ void BufferCache

::UpdateIndexBuffer() { (index_buffer_ref.count + index_buffer_ref.first) * index_buffer_ref.FormatSizeInBytes(); const u32 size = std::min(address_size, draw_size); if (size == 0 || !cpu_addr) { - index_buffer = NULL_BINDING; + channel_state->index_buffer = NULL_BINDING; return; } - index_buffer = Binding{ + channel_state->index_buffer = Binding{ .cpu_addr = *cpu_addr, .size = size, .buffer_id = FindBuffer(*cpu_addr, size), @@ -1051,13 +1062,13 @@ void BufferCache

::UpdateVertexBuffer(u32 index) { const u32 address_size = static_cast(gpu_addr_end - gpu_addr_begin); u32 size = address_size; // TODO: Analyze stride and number of vertices if (array.enable == 0 || size == 0 || !cpu_addr) { - vertex_buffers[index] = NULL_BINDING; + channel_state->vertex_buffers[index] = NULL_BINDING; return; } if (!gpu_memory->IsWithinGPUAddressRange(gpu_addr_end)) { size = static_cast(gpu_memory->MaxContinuousRange(gpu_addr_begin, size)); } - vertex_buffers[index] = Binding{ + channel_state->vertex_buffers[index] = Binding{ .cpu_addr = *cpu_addr, .size = size, .buffer_id = FindBuffer(*cpu_addr, size), @@ -1079,23 +1090,24 @@ void BufferCache

::UpdateDrawIndirect() { }; }; if (current_draw_indirect->include_count) { - update(current_draw_indirect->count_start_address, sizeof(u32), count_buffer_binding); + update(current_draw_indirect->count_start_address, sizeof(u32), + channel_state->count_buffer_binding); } update(current_draw_indirect->indirect_start_address, current_draw_indirect->buffer_size, - indirect_buffer_binding); + channel_state->indirect_buffer_binding); } template void BufferCache

::UpdateUniformBuffers(size_t stage) { - ForEachEnabledBit(enabled_uniform_buffer_masks[stage], [&](u32 index) { - Binding& binding = uniform_buffers[stage][index]; + ForEachEnabledBit(channel_state->enabled_uniform_buffer_masks[stage], [&](u32 index) { + Binding& binding = channel_state->uniform_buffers[stage][index]; if (binding.buffer_id) { // Already updated return; } // Mark as dirty if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { - dirty_uniform_buffers[stage] |= 1U << index; + channel_state->dirty_uniform_buffers[stage] |= 1U << index; } // Resolve buffer binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size); @@ -1104,10 +1116,10 @@ void BufferCache

::UpdateUniformBuffers(size_t stage) { template void BufferCache

::UpdateStorageBuffers(size_t stage) { - const u32 written_mask = written_storage_buffers[stage]; - ForEachEnabledBit(enabled_storage_buffers[stage], [&](u32 index) { + const u32 written_mask = channel_state->written_storage_buffers[stage]; + ForEachEnabledBit(channel_state->enabled_storage_buffers[stage], [&](u32 index) { // Resolve buffer - Binding& binding = storage_buffers[stage][index]; + Binding& binding = channel_state->storage_buffers[stage][index]; const BufferId buffer_id = FindBuffer(binding.cpu_addr, binding.size); binding.buffer_id = buffer_id; // Mark buffer as written if needed @@ -1119,11 +1131,11 @@ void BufferCache

::UpdateStorageBuffers(size_t stage) { template void BufferCache

::UpdateTextureBuffers(size_t stage) { - ForEachEnabledBit(enabled_texture_buffers[stage], [&](u32 index) { - Binding& binding = texture_buffers[stage][index]; + ForEachEnabledBit(channel_state->enabled_texture_buffers[stage], [&](u32 index) { + Binding& binding = channel_state->texture_buffers[stage][index]; binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size); // Mark buffer as written if needed - if (((written_texture_buffers[stage] >> index) & 1) != 0) { + if (((channel_state->written_texture_buffers[stage] >> index) & 1) != 0) { MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size); } }); @@ -1146,11 +1158,11 @@ void BufferCache

::UpdateTransformFeedbackBuffer(u32 index) { const u32 size = binding.size; const std::optional cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); if (binding.enable == 0 || size == 0 || !cpu_addr) { - transform_feedback_buffers[index] = NULL_BINDING; + channel_state->transform_feedback_buffers[index] = NULL_BINDING; return; } const BufferId buffer_id = FindBuffer(*cpu_addr, size); - transform_feedback_buffers[index] = Binding{ + channel_state->transform_feedback_buffers[index] = Binding{ .cpu_addr = *cpu_addr, .size = size, .buffer_id = buffer_id, @@ -1160,8 +1172,8 @@ void BufferCache

::UpdateTransformFeedbackBuffer(u32 index) { template void BufferCache

::UpdateComputeUniformBuffers() { - ForEachEnabledBit(enabled_compute_uniform_buffer_mask, [&](u32 index) { - Binding& binding = compute_uniform_buffers[index]; + ForEachEnabledBit(channel_state->enabled_compute_uniform_buffer_mask, [&](u32 index) { + Binding& binding = channel_state->compute_uniform_buffers[index]; binding = NULL_BINDING; const auto& launch_desc = kepler_compute->launch_description; if (((launch_desc.const_buffer_enable_mask >> index) & 1) != 0) { @@ -1178,12 +1190,12 @@ void BufferCache

::UpdateComputeUniformBuffers() { template void BufferCache

::UpdateComputeStorageBuffers() { - ForEachEnabledBit(enabled_compute_storage_buffers, [&](u32 index) { + ForEachEnabledBit(channel_state->enabled_compute_storage_buffers, [&](u32 index) { // Resolve buffer - Binding& binding = compute_storage_buffers[index]; + Binding& binding = channel_state->compute_storage_buffers[index]; binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size); // Mark as written if needed - if (((written_compute_storage_buffers >> index) & 1) != 0) { + if (((channel_state->written_compute_storage_buffers >> index) & 1) != 0) { MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size); } }); @@ -1191,11 +1203,11 @@ void BufferCache

::UpdateComputeStorageBuffers() { template void BufferCache

::UpdateComputeTextureBuffers() { - ForEachEnabledBit(enabled_compute_texture_buffers, [&](u32 index) { - Binding& binding = compute_texture_buffers[index]; + ForEachEnabledBit(channel_state->enabled_compute_texture_buffers, [&](u32 index) { + Binding& binding = channel_state->compute_texture_buffers[index]; binding.buffer_id = FindBuffer(binding.cpu_addr, binding.size); // Mark as written if needed - if (((written_compute_texture_buffers >> index) & 1) != 0) { + if (((channel_state->written_compute_texture_buffers >> index) & 1) != 0) { MarkWrittenBuffer(binding.buffer_id, binding.cpu_addr, binding.size); } }); @@ -1610,13 +1622,13 @@ void BufferCache

::DeleteBuffer(BufferId buffer_id, bool do_not_mark) { const auto replace = [scalar_replace](std::span bindings) { std::ranges::for_each(bindings, scalar_replace); }; - scalar_replace(index_buffer); - replace(vertex_buffers); - std::ranges::for_each(uniform_buffers, replace); - std::ranges::for_each(storage_buffers, replace); - replace(transform_feedback_buffers); - replace(compute_uniform_buffers); - replace(compute_storage_buffers); + scalar_replace(channel_state->index_buffer); + replace(channel_state->vertex_buffers); + std::ranges::for_each(channel_state->uniform_buffers, replace); + std::ranges::for_each(channel_state->storage_buffers, replace); + replace(channel_state->transform_feedback_buffers); + replace(channel_state->compute_uniform_buffers); + replace(channel_state->compute_storage_buffers); // Mark the whole buffer as CPU written to stop tracking CPU writes if (!do_not_mark) { @@ -1634,8 +1646,8 @@ void BufferCache

::DeleteBuffer(BufferId buffer_id, bool do_not_mark) { template void BufferCache

::NotifyBufferDeletion() { if constexpr (HAS_PERSISTENT_UNIFORM_BUFFER_BINDINGS) { - dirty_uniform_buffers.fill(~u32{0}); - uniform_buffer_binding_sizes.fill({}); + channel_state->dirty_uniform_buffers.fill(~u32{0}); + channel_state->uniform_buffer_binding_sizes.fill({}); } auto& flags = maxwell3d->dirty.flags; flags[Dirty::IndexBuffer] = true; @@ -1643,13 +1655,12 @@ void BufferCache

::NotifyBufferDeletion() { for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) { flags[Dirty::VertexBuffer0 + index] = true; } - has_deleted_buffers = true; + channel_state->has_deleted_buffers = true; } template -typename BufferCache

::Binding BufferCache

::StorageBufferBinding(GPUVAddr ssbo_addr, - u32 cbuf_index, - bool is_written) const { +Binding BufferCache

::StorageBufferBinding(GPUVAddr ssbo_addr, u32 cbuf_index, + bool is_written) const { const GPUVAddr gpu_addr = gpu_memory->Read(ssbo_addr); const auto size = [&]() { const bool is_nvn_cbuf = cbuf_index == 0; @@ -1681,8 +1692,8 @@ typename BufferCache

::Binding BufferCache

::StorageBufferBinding(GPUVAddr s } template -typename BufferCache

::TextureBufferBinding BufferCache

::GetTextureBufferBinding( - GPUVAddr gpu_addr, u32 size, PixelFormat format) { +TextureBufferBinding BufferCache

::GetTextureBufferBinding(GPUVAddr gpu_addr, u32 size, + PixelFormat format) { const std::optional cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr); TextureBufferBinding binding; if (!cpu_addr || size == 0) { @@ -1721,7 +1732,7 @@ 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 ((fast_bound_uniform_buffers[stage] >> binding_index) & 1) != 0; + return ((channel_state->fast_bound_uniform_buffers[stage] >> binding_index) & 1) != 0; } else { // Only OpenGL has fast uniform buffers return false; @@ -1730,14 +1741,14 @@ bool BufferCache

::HasFastUniformBufferBound(size_t stage, u32 binding_index) template std::pair::Buffer*, u32> BufferCache

::GetDrawIndirectCount() { - auto& buffer = slot_buffers[count_buffer_binding.buffer_id]; - return std::make_pair(&buffer, buffer.Offset(count_buffer_binding.cpu_addr)); + auto& buffer = slot_buffers[channel_state->count_buffer_binding.buffer_id]; + return std::make_pair(&buffer, buffer.Offset(channel_state->count_buffer_binding.cpu_addr)); } template std::pair::Buffer*, u32> BufferCache

::GetDrawIndirectBuffer() { - auto& buffer = slot_buffers[indirect_buffer_binding.buffer_id]; - return std::make_pair(&buffer, buffer.Offset(indirect_buffer_binding.cpu_addr)); + auto& buffer = slot_buffers[channel_state->indirect_buffer_binding.buffer_id]; + return std::make_pair(&buffer, buffer.Offset(channel_state->indirect_buffer_binding.cpu_addr)); } } // namespace VideoCommon diff --git a/src/video_core/buffer_cache/buffer_cache_base.h b/src/video_core/buffer_cache/buffer_cache_base.h index ac00d4d9d..c689fe06b 100644 --- a/src/video_core/buffer_cache/buffer_cache_base.h +++ b/src/video_core/buffer_cache/buffer_cache_base.h @@ -86,8 +86,78 @@ enum class ObtainBufferOperation : u32 { MarkQuery = 3, }; -template -class BufferCache : public VideoCommon::ChannelSetupCaches { +static constexpr BufferId NULL_BUFFER_ID{0}; +static constexpr u32 DEFAULT_SKIP_CACHE_SIZE = static_cast(4_KiB); + +struct Binding { + VAddr cpu_addr{}; + u32 size{}; + BufferId buffer_id; +}; + +struct TextureBufferBinding : Binding { + PixelFormat format; +}; + +static constexpr Binding NULL_BINDING{ + .cpu_addr = 0, + .size = 0, + .buffer_id = NULL_BUFFER_ID, +}; + +class BufferCacheChannelInfo : public ChannelInfo { +public: + BufferCacheChannelInfo() = delete; + BufferCacheChannelInfo(Tegra::Control::ChannelState& state) noexcept : ChannelInfo(state) {} + BufferCacheChannelInfo(const BufferCacheChannelInfo& state) = delete; + BufferCacheChannelInfo& operator=(const BufferCacheChannelInfo&) = delete; + + Binding index_buffer; + std::array vertex_buffers; + std::array, NUM_STAGES> uniform_buffers; + std::array, NUM_STAGES> storage_buffers; + std::array, NUM_STAGES> texture_buffers; + std::array transform_feedback_buffers; + Binding count_buffer_binding; + Binding indirect_buffer_binding; + + std::array compute_uniform_buffers; + std::array compute_storage_buffers; + std::array compute_texture_buffers; + + std::array enabled_uniform_buffer_masks{}; + u32 enabled_compute_uniform_buffer_mask = 0; + + const UniformBufferSizes* uniform_buffer_sizes{}; + const ComputeUniformBufferSizes* compute_uniform_buffer_sizes{}; + + std::array enabled_storage_buffers{}; + std::array written_storage_buffers{}; + u32 enabled_compute_storage_buffers = 0; + u32 written_compute_storage_buffers = 0; + + std::array enabled_texture_buffers{}; + std::array written_texture_buffers{}; + std::array image_texture_buffers{}; + u32 enabled_compute_texture_buffers = 0; + u32 written_compute_texture_buffers = 0; + u32 image_compute_texture_buffers = 0; + + std::array uniform_cache_hits{}; + std::array uniform_cache_shots{}; + + u32 uniform_buffer_skip_cache_size = DEFAULT_SKIP_CACHE_SIZE; + + bool has_deleted_buffers = false; + + std::array dirty_uniform_buffers{}; + std::array fast_bound_uniform_buffers{}; + std::array, NUM_STAGES> + uniform_buffer_binding_sizes{}; +}; + +template +class BufferCache : public VideoCommon::ChannelSetupCaches { // Page size for caching purposes. // This is unrelated to the CPU page size and it can be changed as it seems optimal. static constexpr u32 CACHING_PAGEBITS = 16; @@ -104,8 +174,6 @@ class BufferCache : public VideoCommon::ChannelSetupCaches; using OverlapCounter = boost::icl::split_interval_map; - struct Empty {}; - struct OverlapResult { std::vector ids; VAddr begin; @@ -158,25 +224,7 @@ class BufferCache : public VideoCommon::ChannelSetupCaches(4_KiB); - explicit BufferCache(VideoCore::RasterizerInterface& rasterizer_, Core::Memory::Memory& cpu_memory_, Runtime& runtime_); @@ -496,51 +544,6 @@ private: u32 last_index_count = 0; - Binding index_buffer; - std::array vertex_buffers; - std::array, NUM_STAGES> uniform_buffers; - std::array, NUM_STAGES> storage_buffers; - std::array, NUM_STAGES> texture_buffers; - std::array transform_feedback_buffers; - Binding count_buffer_binding; - Binding indirect_buffer_binding; - - std::array compute_uniform_buffers; - std::array compute_storage_buffers; - std::array compute_texture_buffers; - - std::array enabled_uniform_buffer_masks{}; - u32 enabled_compute_uniform_buffer_mask = 0; - - const UniformBufferSizes* uniform_buffer_sizes{}; - const ComputeUniformBufferSizes* compute_uniform_buffer_sizes{}; - - std::array enabled_storage_buffers{}; - std::array written_storage_buffers{}; - u32 enabled_compute_storage_buffers = 0; - u32 written_compute_storage_buffers = 0; - - std::array enabled_texture_buffers{}; - std::array written_texture_buffers{}; - std::array image_texture_buffers{}; - u32 enabled_compute_texture_buffers = 0; - u32 written_compute_texture_buffers = 0; - u32 image_compute_texture_buffers = 0; - - std::array uniform_cache_hits{}; - std::array uniform_cache_shots{}; - - u32 uniform_buffer_skip_cache_size = DEFAULT_SKIP_CACHE_SIZE; - - bool has_deleted_buffers = false; - - std::conditional_t, Empty> - dirty_uniform_buffers{}; - std::conditional_t, Empty> fast_bound_uniform_buffers{}; - std::conditional_t, NUM_STAGES>, Empty> - uniform_buffer_binding_sizes{}; - MemoryTracker memory_tracker; IntervalSet uncommitted_ranges; IntervalSet common_ranges; diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index 6af4ae793..6d3bda192 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -117,7 +117,7 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_) for (auto& stage_uniforms : fast_uniforms) { for (OGLBuffer& buffer : stage_uniforms) { buffer.Create(); - glNamedBufferData(buffer.handle, BufferCache::DEFAULT_SKIP_CACHE_SIZE, nullptr, + glNamedBufferData(buffer.handle, VideoCommon::DEFAULT_SKIP_CACHE_SIZE, nullptr, GL_STREAM_DRAW); } }