当前位置:首页 > 报告详情

把《杀手》装进口袋.pdf

上传人: 柒柒 编号:1099639 2026-01-25 47页 6.97MB

1、Scope Hitman&G2 Internal port Live updates-Elusive Targets Older branch of G2 No content changes No geometry streaming system Aspire to have low maintenance costMobile Platforms Smartphones&Tablets System on Chip(SoC)design Unified Memory 3-6W Power Budget Peak performance is not sustainable RAM Acc

2、ess is energy intensiveo1W=150-200MB 60fpsGroundwork Game Porting ToolkitoTrying out a Windows-DX12 build on macOS Try to find early failure points oMemory BudgetoUnsupported FeaturesTarget Min Spec oiPhone 15 Pro(A17Pro)o8GB unified memory,6GB usable iOS 18.0-“Metal3oArgument Buffers-Tier 2oResiden

3、cy Sets*Device Capabilitiesoiphone-performance-gaming-tier Entitlements:oincreased-memory-limitosustained-executionRender Thread Main thread deals with gameplay logic and loading Render thread culls and records GPU commandsoReflect step copies necessary data from main thread to render thread Configu

4、rable maximum latency between main and renderoUsed for dealing with main thread spikes Not 1:1 with display framesoRender thread takes over during loading&sometimes doesnt present a frameWhose Main Thread Is It Anyway?Apple APIs are based on event driven model A lot of them can only run on main thre

5、adoRuns an event loop for processing I/O and callbacksoAvoid blocking as much as possible Our main and render threads have their own event loopsoNeeds to do blocking waits for other threads Game runs on its own“main”threadoPolls CFRunLoop to allow system main thread to schedule functionoSchedules an

6、 update function on system main thread every frameAcquiring a Drawable Metal term for back buffer Various options availableonextDrawable,CADisplayLink,CAMetalDisplayLink Threading setup makes things difficult Consistency problemsoExperienced up to 2 ms variabilityoProblematic when GPU downclocks to

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
1. **目标平台与配置**:适配iPhone 15 Pro(A17 Pro,8GB统一内存,可用~6GB),iOS 18.0支持Metal3(Argument Buffers Tier 2、Residency Sets)。 2. **渲染优化**:双线程架构(主线程逻辑+渲染线程GPU指令),可配置延迟;使用`presentAfterMinimumDuration`统一帧 pacing。 3. **资源管理**:纹理采用ASTC压缩;缓冲区分只读(大MTLBuffer子分配)与可写(独立分配);纹理堆(MTLHeap)平衡 residency 调用与分配。 4. **性能挑战**:峰值性能不可持续(RAM访问1W=150-200MB@60fps);GPU降频导致帧时间波动(最高2ms);异步计算未获收益。 5. **工具链**:HLSL经Metal Shader Converter转MSL;PSO预缓存提升加载速度;Thin LTO显著优化性能。 6. **成果**:30 FPS运行于iPhone 15 Pro及更新机型,强调低维护成本与跨平台移植(Windows-DX12至macOS)。
移植难点? 性能优化? Metal技巧?
客服
商务合作
小程序
服务号
折叠