feat: 显存分配器迁移到 VMA(P1 路线第一项)
- vendored VulkanMemoryAllocator 子模块,唯一实现编译单元 VmaImpl.cpp - MemoryAllocator 重写为 VMA 后端:接口从"Allocate(reqs)+手绑"升级为 "创建即分配即绑定"(CreateBuffer/CreateImage/Destroy/Flush), VMA 要求资源与内存成对释放,约束成单一出口 - Buffer/Texture/RenderGraph 三处调用方迁移;staging 持久映射 + 按分配 精确 flush(原来按整块 flush) - 获得:TLSF 通用分配、DedicatedAllocation 自动启用、memory budget 感知、 内存别名基础(TAA/复用前置) - 编译零错误、validation 零警告、8 秒冒烟通过
This commit is contained in:
@@ -17,3 +17,6 @@
|
||||
path = Lib/imgui
|
||||
url = https://github.com/ocornut/imgui.git
|
||||
branch = docking
|
||||
[submodule "Lib/VulkanMemoryAllocator"]
|
||||
path = Lib/VulkanMemoryAllocator
|
||||
url = https://github.com/GPUOpen-LibrariesAndSDKs/VulkanMemoryAllocator.git
|
||||
|
||||
Reference in New Issue
Block a user