Optimize RmlUI renderer to use one vertex and index buffer per frame. (#518)

This commit is contained in:
Darío 2024-12-23 00:57:59 -03:00 committed by GitHub
parent 77ec5b5b4b
commit bed19044da
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -113,6 +113,14 @@ T from_bytes_le(const char* input) {
void load_document();
class RmlRenderInterface_RT64 : public Rml::RenderInterfaceCompatibility {
struct DynamicBuffer {
std::unique_ptr<RT64::RenderBuffer> buffer_{};
uint32_t size_ = 0;
uint32_t bytes_used_ = 0;
uint8_t* mapped_data_ = nullptr;
RT64::RenderBufferFlags flags_ = RT64::RenderBufferFlag::NONE;
};
static constexpr uint32_t per_frame_descriptor_set = 0;
static constexpr uint32_t per_draw_descriptor_set = 1;
@ -137,9 +145,9 @@ class RmlRenderInterface_RT64 : public Rml::RenderInterfaceCompatibility {
Rml::Matrix4f mvp_ = Rml::Matrix4f::Identity();
std::unordered_map<Rml::TextureHandle, TextureHandle> textures_{};
Rml::TextureHandle texture_count_ = 1; // Start at 1 to reserve texture 0 as the 1x1 pixel white texture
std::unique_ptr<RT64::RenderBuffer> upload_buffer_{};
std::unique_ptr<RT64::RenderBuffer> vertex_buffer_{};
std::unique_ptr<RT64::RenderBuffer> index_buffer_{};
DynamicBuffer upload_buffer_;
DynamicBuffer vertex_buffer_;
DynamicBuffer index_buffer_;
std::unique_ptr<RT64::RenderSampler> nearestSampler_{};
std::unique_ptr<RT64::RenderSampler> linearSampler_{};
std::unique_ptr<RT64::RenderShader> vertex_shader_{};
@ -155,11 +163,6 @@ class RmlRenderInterface_RT64 : public Rml::RenderInterfaceCompatibility {
std::unique_ptr<RT64::RenderDescriptorSet> screen_descriptor_set_{};
std::unique_ptr<RT64::RenderBuffer> screen_vertex_buffer_{};
uint64_t screen_vertex_buffer_size_ = 0;
uint32_t upload_buffer_size_ = 0;
uint32_t upload_buffer_bytes_used_ = 0;
uint8_t* upload_buffer_mapped_data_ = nullptr;
uint32_t vertex_buffer_size_ = 0;
uint32_t index_buffer_size_ = 0;
uint32_t gTexture_descriptor_index;
RT64::RenderInputSlot vertex_slot_{ 0, sizeof(Rml::Vertex) };
RT64::RenderCommandList* list_ = nullptr;
@ -175,10 +178,13 @@ public:
multisampling_.sampleCount = desired_sample_count;
}
vertex_buffer_.flags_ = RT64::RenderBufferFlag::VERTEX;
index_buffer_.flags_ = RT64::RenderBufferFlag::INDEX;
// Create the texture upload buffer, vertex buffer and index buffer
resize_upload_buffer(initial_upload_buffer_size, false);
resize_vertex_buffer(initial_vertex_buffer_size);
resize_index_buffer(initial_index_buffer_size);
resize_dynamic_buffer(upload_buffer_, initial_upload_buffer_size, false);
resize_dynamic_buffer(vertex_buffer_, initial_vertex_buffer_size, false);
resize_dynamic_buffer(index_buffer_, initial_index_buffer_size, false);
// Describe the vertex format
std::vector<RT64::RenderInputElement> vertex_elements{};
@ -268,90 +274,78 @@ public:
}
}
void resize_upload_buffer(uint32_t new_size, bool map = true) {
// Unmap the upload buffer if it's mapped
if (upload_buffer_mapped_data_ != nullptr) {
upload_buffer_->unmap();
void reset_dynamic_buffer(DynamicBuffer &dynamic_buffer) {
assert(dynamic_buffer.mapped_data_ == nullptr);
dynamic_buffer.bytes_used_ = 0;
dynamic_buffer.mapped_data_ = reinterpret_cast<uint8_t*>(dynamic_buffer.buffer_->map());
}
// If there's already an upload buffer, move it into the stale buffers so it persists until the start of next frame.
if (upload_buffer_) {
stale_buffers_.emplace_back(std::move(upload_buffer_));
void end_dynamic_buffer(DynamicBuffer &dynamic_buffer) {
assert(dynamic_buffer.mapped_data_ != nullptr);
dynamic_buffer.buffer_->unmap();
dynamic_buffer.mapped_data_ = nullptr;
}
// Create the new upload buffer, update the size and map it.
upload_buffer_ = render_context_->device->createBuffer(RT64::RenderBufferDesc::UploadBuffer(new_size));
upload_buffer_size_ = new_size;
upload_buffer_bytes_used_ = 0;
void resize_dynamic_buffer(DynamicBuffer &dynamic_buffer, uint32_t new_size, bool map = true) {
// Unmap the buffer if it's mapped
if (dynamic_buffer.mapped_data_ != nullptr) {
dynamic_buffer.buffer_->unmap();
}
// If there's already a buffer, move it into the stale buffers so it persists until the start of next frame.
if (dynamic_buffer.buffer_ != nullptr) {
stale_buffers_.emplace_back(std::move(dynamic_buffer.buffer_));
}
// Create the new buffer, update the size and map it.
dynamic_buffer.buffer_ = render_context_->device->createBuffer(RT64::RenderBufferDesc::UploadBuffer(new_size, dynamic_buffer.flags_));
dynamic_buffer.size_ = new_size;
dynamic_buffer.bytes_used_ = 0;
if (map) {
upload_buffer_mapped_data_ = reinterpret_cast<uint8_t*>(upload_buffer_->map());
}
else {
upload_buffer_mapped_data_ = nullptr;
dynamic_buffer.mapped_data_ = reinterpret_cast<uint8_t*>(dynamic_buffer.buffer_->map());
}
}
uint32_t allocate_upload_data(uint32_t num_bytes) {
// Check if there's enough remaining room in the upload buffer to allocate the requested bytes.
uint32_t total_bytes = num_bytes + upload_buffer_bytes_used_;
uint32_t allocate_dynamic_data(DynamicBuffer &dynamic_buffer, uint32_t num_bytes) {
// Check if there's enough remaining room in the buffer to allocate the requested bytes.
uint32_t total_bytes = num_bytes + dynamic_buffer.bytes_used_;
if (total_bytes > upload_buffer_size_) {
// There isn't, so mark the current upload buffer as stale and allocate a new one with 50% more space than the required amount.
resize_upload_buffer(total_bytes + total_bytes / 2);
if (total_bytes > dynamic_buffer.size_) {
// There isn't, so mark the current buffer as stale and allocate a new one with 50% more space than the required amount.
resize_dynamic_buffer(dynamic_buffer, total_bytes + total_bytes / 2);
}
// Record the current end of the upload buffer to return.
uint32_t offset = upload_buffer_bytes_used_;
// Record the current end of the buffer to return.
uint32_t offset = dynamic_buffer.bytes_used_;
// Bump the upload buffer's end forward by the number of bytes allocated.
upload_buffer_bytes_used_ += num_bytes;
// Bump the buffer's end forward by the number of bytes allocated.
dynamic_buffer.bytes_used_ += num_bytes;
return offset;
}
uint32_t allocate_upload_data_aligned(uint32_t num_bytes, uint32_t alignment) {
// Check if there's enough remaining room in the upload buffer to allocate the requested bytes.
uint32_t total_bytes = num_bytes + upload_buffer_bytes_used_;
uint32_t allocate_dynamic_data_aligned(DynamicBuffer &dynamic_buffer, uint32_t num_bytes, uint32_t alignment) {
// Check if there's enough remaining room in the buffer to allocate the requested bytes.
uint32_t total_bytes = num_bytes + dynamic_buffer.bytes_used_;
// Determine the amount of padding needed to meet the target alignment.
uint32_t padding_bytes = ((upload_buffer_bytes_used_ + alignment - 1) / alignment) * alignment - upload_buffer_bytes_used_;
uint32_t padding_bytes = ((dynamic_buffer.bytes_used_ + alignment - 1) / alignment) * alignment - dynamic_buffer.bytes_used_;
// If there isn't enough room to allocate the required bytes plus the padding then resize the upload buffer and allocate from the start of the new one.
if (total_bytes + padding_bytes > upload_buffer_size_) {
resize_upload_buffer(total_bytes + total_bytes / 2);
// If there isn't enough room to allocate the required bytes plus the padding then resize the buffer and allocate from the start of the new one.
if (total_bytes + padding_bytes > dynamic_buffer.size_) {
resize_dynamic_buffer(dynamic_buffer, total_bytes + total_bytes / 2);
upload_buffer_bytes_used_ += num_bytes;
dynamic_buffer.bytes_used_ += num_bytes;
return 0;
}
// Otherwise allocate the padding and required bytes and offset the allocated position by the padding size.
return allocate_upload_data(padding_bytes + num_bytes) + padding_bytes;
}
void resize_vertex_buffer(uint32_t new_size) {
if (vertex_buffer_) {
stale_buffers_.emplace_back(std::move(vertex_buffer_));
}
vertex_buffer_ = render_context_->device->createBuffer(RT64::RenderBufferDesc::VertexBuffer(new_size, RT64::RenderHeapType::DEFAULT));
vertex_buffer_size_ = new_size;
}
void resize_index_buffer(uint32_t new_size) {
if (index_buffer_) {
stale_buffers_.emplace_back(std::move(index_buffer_));
}
index_buffer_ = render_context_->device->createBuffer(RT64::RenderBufferDesc::IndexBuffer(new_size, RT64::RenderHeapType::DEFAULT));
index_buffer_size_ = new_size;
return allocate_dynamic_data(dynamic_buffer, padding_bytes + num_bytes) + padding_bytes;
}
void RenderGeometry(Rml::Vertex* vertices, int num_vertices, int* indices, int num_indices, Rml::TextureHandle texture, const Rml::Vector2f& translation) override {
uint32_t vert_size_bytes = num_vertices * sizeof(*vertices);
uint32_t index_size_bytes = num_indices * sizeof(*indices);
uint32_t total_bytes = vert_size_bytes + index_size_bytes;
uint32_t index_bytes_start = vert_size_bytes;
if (!textures_.contains(texture)) {
if (texture == 0) {
// Create a 1x1 pixel white texture as the first handle
@ -363,37 +357,13 @@ public:
}
}
uint32_t upload_buffer_offset = allocate_upload_data(total_bytes);
if (vert_size_bytes > vertex_buffer_size_) {
resize_vertex_buffer(vert_size_bytes + vert_size_bytes / 2);
}
if (index_size_bytes > index_buffer_size_) {
resize_index_buffer(index_size_bytes + index_size_bytes / 2);
}
// Copy the vertex and index data into the mapped upload buffer.
memcpy(upload_buffer_mapped_data_ + upload_buffer_offset, vertices, vert_size_bytes);
memcpy(upload_buffer_mapped_data_ + upload_buffer_offset + vert_size_bytes, indices, index_size_bytes);
// Prepare the vertex and index buffers for being copied to.
RT64::RenderBufferBarrier copy_barriers[] = {
RT64::RenderBufferBarrier(vertex_buffer_.get(), RT64::RenderBufferAccess::WRITE),
RT64::RenderBufferBarrier(index_buffer_.get(), RT64::RenderBufferAccess::WRITE)
};
list_->barriers(RT64::RenderBarrierStage::COPY, copy_barriers, uint32_t(std::size(copy_barriers)));
// Copy from the upload buffer to the vertex and index buffers.
list_->copyBufferRegion(vertex_buffer_->at(0), upload_buffer_->at(upload_buffer_offset), vert_size_bytes);
list_->copyBufferRegion(index_buffer_->at(0), upload_buffer_->at(upload_buffer_offset + index_bytes_start), index_size_bytes);
// Prepare the vertex and index buffers for being used for rendering.
RT64::RenderBufferBarrier usage_barriers[] = {
RT64::RenderBufferBarrier(vertex_buffer_.get(), RT64::RenderBufferAccess::READ),
RT64::RenderBufferBarrier(index_buffer_.get(), RT64::RenderBufferAccess::READ)
};
list_->barriers(RT64::RenderBarrierStage::GRAPHICS, usage_barriers, uint32_t(std::size(usage_barriers)));
// Copy the vertex and index data into the mapped buffers.
uint32_t vert_size_bytes = num_vertices * sizeof(*vertices);
uint32_t index_size_bytes = num_indices * sizeof(*indices);
uint32_t vertex_buffer_offset = allocate_dynamic_data(vertex_buffer_, vert_size_bytes);
uint32_t index_buffer_offset = allocate_dynamic_data(index_buffer_, index_size_bytes);
memcpy(vertex_buffer_.mapped_data_ + vertex_buffer_offset, vertices, vert_size_bytes);
memcpy(index_buffer_.mapped_data_ + index_buffer_offset, indices, index_size_bytes);
list_->setViewports(RT64::RenderViewport{ 0, 0, float(window_width_), float(window_height_) });
if (scissor_enabled_) {
@ -407,9 +377,9 @@ public:
list_->setScissors(RT64::RenderRect{ 0, 0, window_width_, window_height_ });
}
RT64::RenderIndexBufferView index_view{index_buffer_->at(0), index_size_bytes, RT64::RenderFormat::R32_UINT};
RT64::RenderIndexBufferView index_view{index_buffer_.buffer_->at(index_buffer_offset), index_size_bytes, RT64::RenderFormat::R32_UINT};
list_->setIndexBuffer(&index_view);
RT64::RenderVertexBufferView vertex_view{vertex_buffer_->at(0), vert_size_bytes};
RT64::RenderVertexBufferView vertex_view{vertex_buffer_.buffer_->at(vertex_buffer_offset), vert_size_bytes};
list_->setVertexBuffers(0, &vertex_view, 1, &vertex_slot_);
list_->setGraphicsDescriptorSet(textures_.at(texture).set.get(), 1);
@ -530,10 +500,10 @@ public:
uint32_t uploaded_size_bytes = row_byte_width * source_dimensions.y;
// Allocate room in the upload buffer for the uploaded data.
uint32_t upload_buffer_offset = allocate_upload_data_aligned(uploaded_size_bytes, 512);
uint32_t upload_buffer_offset = allocate_dynamic_data_aligned(upload_buffer_, uploaded_size_bytes, 512);
// Copy the source data into the upload buffer.
uint8_t* dst_data = upload_buffer_mapped_data_ + upload_buffer_offset;
uint8_t* dst_data = upload_buffer_.mapped_data_ + upload_buffer_offset;
if (row_byte_padding == 0) {
// Copy row-by-row if the image is flipped.
@ -567,7 +537,7 @@ public:
// Copy the upload buffer into the texture.
list_->copyTextureRegion(
RT64::RenderTextureCopyLocation::Subresource(texture.get()),
RT64::RenderTextureCopyLocation::PlacedFootprint(upload_buffer_.get(), RmlTextureFormat, source_dimensions.x, source_dimensions.y, 1, row_width, upload_buffer_offset));
RT64::RenderTextureCopyLocation::PlacedFootprint(upload_buffer_.buffer_.get(), RmlTextureFormat, source_dimensions.x, source_dimensions.y, 1, row_width, upload_buffer_offset));
// Prepare the texture for being read from a pixel shader.
list_->barriers(RT64::RenderBarrierStage::GRAPHICS, RT64::RenderTextureBarrier(texture.get(), RT64::RenderTextureLayout::SHADER_READ));
@ -631,9 +601,10 @@ public:
// Clear out any stale buffers from the last command list.
stale_buffers_.clear();
// Reset and map the upload buffer.
upload_buffer_bytes_used_ = 0;
upload_buffer_mapped_data_ = reinterpret_cast<uint8_t*>(upload_buffer_->map());
// Reset buffers.
reset_dynamic_buffer(upload_buffer_);
reset_dynamic_buffer(vertex_buffer_);
reset_dynamic_buffer(index_buffer_);
// Set an internal texture as the render target if MSAA is enabled.
if (multisampling_.sampleCount > 1) {
@ -671,13 +642,11 @@ public:
list->drawInstanced(3, 1, 0, 0);
}
list_ = nullptr;
end_dynamic_buffer(upload_buffer_);
end_dynamic_buffer(vertex_buffer_);
end_dynamic_buffer(index_buffer_);
// Unmap the upload buffer if it's mapped.
if (upload_buffer_mapped_data_) {
upload_buffer_->unmap();
upload_buffer_mapped_data_ = nullptr;
}
list_ = nullptr;
}
};