Stowage/godot
2
0
Fork 0
mirror of https://github.com/godotengine/godot.git synced 2025-12-08 06:09:55 +00:00
Code Issues Projects Releases Packages Wiki Activity

2D: Switch to VBOs for instance data

Browse source 
- Add support for vertex bindings and UMA vertex buffers in D3D12.
- Simplify 2D instance params and move more into per-batch data to save
  bandwidth

Co-authored-by: Skyth <19259897+blueskythlikesclouds@users.noreply.github.com>
Co-authored-by: Clay John <claynjohn@gmail.com>
Co-authored-by: A Thousand Ships <96648715+athousandships@users.noreply.github.com>
This commit is contained in:
Stuart Carnie 2025-11-07 05:50:02 +11:00
parent bad1287b62
commit 90c0e6acca
25 changed files with 893 additions and 256 deletions
Download patch file Download diff file Expand all files Collapse all files

68
drivers/d3d12/rendering_device_driver_d3d12.cpp
View file

@ -1165,6 +1165,24 @@ uint8_t *RenderingDeviceDriverD3D12::buffer_persistent_map_advance(BufferID p_bu
return buf_info->persistent_ptr + buf_info->frame_idx * buf_info->size;
}
uint64_t RenderingDeviceDriverD3D12::buffer_get_dynamic_offsets(Span<BufferID> p_buffers) {
uint64_t mask = 0u;
uint64_t shift = 0u;
for (const BufferID &buf : p_buffers) {
const BufferInfo *buf_info = (const BufferInfo *)buf.id;
if (!buf_info->is_dynamic()) {
continue;
}
const BufferDynamicInfo *dyn_buf = (const BufferDynamicInfo *)buf.id;
mask |= dyn_buf->frame_idx << shift;
// We can encode the frame index in 2 bits since frame_count won't be > 4.
shift += 2UL;
}
return mask;
}
uint64_t RenderingDeviceDriverD3D12::buffer_get_device_address(BufferID p_buffer) {
const BufferInfo *buf_info = (const BufferInfo *)p_buffer.id;
return buf_info->resource->GetGPUVirtualAddress();
@ -2183,27 +2201,36 @@ bool RenderingDeviceDriverD3D12::sampler_is_format_supported_for_filter(DataForm
/**** VERTEX ARRAY ****/
/**********************/
RDD::VertexFormatID RenderingDeviceDriverD3D12::vertex_format_create(VectorView<VertexAttribute> p_vertex_attribs) {
RDD::VertexFormatID RenderingDeviceDriverD3D12::vertex_format_create(Span<VertexAttribute> p_vertex_attribs, const VertexAttributeBindingsMap &p_vertex_bindings) {
VertexFormatInfo *vf_info = VersatileResource::allocate<VertexFormatInfo>(resources_allocator);
vf_info->input_elem_descs.resize(p_vertex_attribs.size());
vf_info->vertex_buffer_strides.resize(p_vertex_attribs.size());
uint32_t max_binding = 0;
for (uint32_t i = 0; i < p_vertex_attribs.size(); i++) {
vf_info->input_elem_descs[i] = {};
vf_info->input_elem_descs[i].SemanticName = "TEXCOORD";
vf_info->input_elem_descs[i].SemanticIndex = p_vertex_attribs[i].location;
vf_info->input_elem_descs[i].Format = RD_TO_D3D12_FORMAT[p_vertex_attribs[i].format].general_format;
vf_info->input_elem_descs[i].InputSlot = i; // TODO: Can the same slot be used if data comes from the same buffer (regardless format)?
vf_info->input_elem_descs[i].AlignedByteOffset = p_vertex_attribs[i].offset;
if (p_vertex_attribs[i].frequency == VERTEX_FREQUENCY_INSTANCE) {
vf_info->input_elem_descs[i].InputSlotClass = D3D12_INPUT_CLASSIFICATION_PER_INSTANCE_DATA;
vf_info->input_elem_descs[i].InstanceDataStepRate = 1;
D3D12_INPUT_ELEMENT_DESC &input_element_desc = vf_info->input_elem_descs[i];
const VertexAttribute &vertex_attrib = p_vertex_attribs[i];
const VertexAttributeBinding &vertex_binding = p_vertex_bindings[vertex_attrib.binding];
input_element_desc = {};
input_element_desc.SemanticName = "TEXCOORD";
input_element_desc.SemanticIndex = vertex_attrib.location;
input_element_desc.Format = RD_TO_D3D12_FORMAT[vertex_attrib.format].general_format;
input_element_desc.InputSlot = vertex_attrib.binding;
input_element_desc.AlignedByteOffset = vertex_attrib.offset;
if (vertex_binding.frequency == VERTEX_FREQUENCY_INSTANCE) {
input_element_desc.InputSlotClass = D3D12_INPUT_CLASSIFICATION_PER_INSTANCE_DATA;
input_element_desc.InstanceDataStepRate = 1;
} else {
vf_info->input_elem_descs[i].InputSlotClass = D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA;
vf_info->input_elem_descs[i].InstanceDataStepRate = 0;
input_element_desc.InputSlotClass = D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA;
input_element_desc.InstanceDataStepRate = 0;
}
vf_info->vertex_buffer_strides[i] = p_vertex_attribs[i].stride;
max_binding = MAX(max_binding, vertex_attrib.binding + 1);
}
vf_info->vertex_buffer_strides.resize(max_binding);
for (const VertexAttributeBindingsMap::KV &vertex_binding_pair : p_vertex_bindings) {
vf_info->vertex_buffer_strides[vertex_binding_pair.key] = vertex_binding_pair.value.stride;
}
return VertexFormatID(vf_info);
@ -5378,7 +5405,7 @@ void RenderingDeviceDriverD3D12::command_render_draw_indirect_count(CommandBuffe
cmd_buf_info->cmd_list->ExecuteIndirect(indirect_cmd_signatures.draw.Get(), p_max_draw_count, indirect_buf_info->resource, p_offset, count_buf_info->resource, p_count_buffer_offset);
}
void RenderingDeviceDriverD3D12::command_render_bind_vertex_buffers(CommandBufferID p_cmd_buffer, uint32_t p_binding_count, const BufferID *p_buffers, const uint64_t *p_offsets) {
void RenderingDeviceDriverD3D12::command_render_bind_vertex_buffers(CommandBufferID p_cmd_buffer, uint32_t p_binding_count, const BufferID *p_buffers, const uint64_t *p_offsets, uint64_t p_dynamic_offsets) {
CommandBufferInfo *cmd_buf_info = (CommandBufferInfo *)p_cmd_buffer.id;
DEV_ASSERT(cmd_buf_info->render_pass_state.current_subpass != UINT32_MAX);
@ -5390,8 +5417,15 @@ void RenderingDeviceDriverD3D12::command_render_bind_vertex_buffers(CommandBuffe
for (uint32_t i = 0; i < p_binding_count; i++) {
BufferInfo *buffer_info = (BufferInfo *)p_buffers[i].id;
uint32_t dynamic_offset = 0;
if (buffer_info->is_dynamic()) {
uint64_t buffer_frame_idx = p_dynamic_offsets & 0x3; // Assuming max 4 frames.
p_dynamic_offsets >>= 2;
dynamic_offset = buffer_frame_idx * buffer_info->size;
}
cmd_buf_info->render_pass_state.vertex_buffer_views[i] = {};
cmd_buf_info->render_pass_state.vertex_buffer_views[i].BufferLocation = buffer_info->resource->GetGPUVirtualAddress() + p_offsets[i];
cmd_buf_info->render_pass_state.vertex_buffer_views[i].BufferLocation = buffer_info->resource->GetGPUVirtualAddress() + dynamic_offset + p_offsets[i];
cmd_buf_info->render_pass_state.vertex_buffer_views[i].SizeInBytes = buffer_info->size - p_offsets[i];
if (!barrier_capabilities.enhanced_barriers_supported) {
_resource_transition_batch(cmd_buf_info, buffer_info, 0, 1, D3D12_RESOURCE_STATE_VERTEX_AND_CONSTANT_BUFFER);