菜单
Use when porting, compiling, benchmarking, or debugging vintage Mac OS X (10.3 Panther through 10.5 Leopard) games across PowerPC (G3, G4, G5), Intel (Lion), and Apple Silicon. Covers endianness, AltiVec SIMD, Mach-O fat binaries, and legacy OpenGL.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/matthewdeaves/old-mac-quake2 --skill legacy-mac-porting命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Use before sending any message containing a file, a line number, a count, a severity, a commit hash, or a statement that X does Y. Also use before retracting one.
Use when about to write "that is the user's call", "waiting on them", or "I'll ask" -- and when a session asks you to choose between options.
Find and apply the next fps or graphics-quality win for the old-Mac Quake II (yquake2) port across the shared bench fleet (G3/Rage128, G4/GeForce2·Radeon9000·9200, Intel-Lion/GMA950, G5/Radeon9600). Re-runnable, each run profiles one target machine class, forms ONE bottleneck-matched hypothesis, implements it cvar-first (then code, gated behind a cvar so one fat binary auto-tunes per machine), benches it safely, keeps or reverts, records it, and says whether more wins remain. Use whenever the user wants more fps, more graphical features, or asks to "optimise/tune" any old Mac.
正在显示 SKILL.md
| name | legacy-mac-porting |
| description | Use when porting, compiling, benchmarking, or debugging vintage Mac OS X (10.3 Panther through 10.5 Leopard) games across PowerPC (G3, G4, G5), Intel (Lion), and Apple Silicon. Covers endianness, AltiVec SIMD, Mach-O fat binaries, and legacy OpenGL. |
Use this guide when diagnosing compiler errors, endianness bugs, AltiVec crashes, or OpenGL rendering artifacts on Mac OS X 10.3-10.7.
| Arch Slice | CPU Family | Machines in Fleet | AltiVec SIMD | Notes |
|---|---|---|---|---|
ppc750 | PowerPC G3 | yosemite (iMac G3 / B&W) | NO | Never pass -maltivec. Max OS 10.3.9 or 10.4.11. |
ppc7400 / ppc7450 | PowerPC G4 | mini-g4, quicksilver, sawtooth | YES (-maltivec) | Vector data must be 16-byte aligned. |
ppc970 | PowerPC G5 | imac-g5, g5-desktop (Dual 2.7) | YES | 64-bit capable, run as 32-bit Mach-O on 10.4/10.5. |
i386 | Intel 32-bit | mini-intel, mini-intel2 (Lion 10.7.5) | SSE2 | Core 2 Duo / Core Duo. |
x86_64 | Intel 64-bit | mini-sl (Snow Leopard 10.6.8) | SSE3/SSE4 | GeForce 9400M / 64-bit Intel. |
arm64 | Apple Silicon | Modern macOS 11-15+ | NEON | Host workstation native slice. |
__BIG_ENDIAN__ vs __LITTLE_ENDIAN__)PowerPC is Big-Endian; x86 and arm64 are Little-Endian.
#if defined(__BIG_ENDIAN__) || defined(__ppc__) || defined(__POWERPC__)
uint32_t val = LittleLong(raw_val);
float fval = LittleFloat(raw_fval);
#endif
htons/htonl).vector float, vector unsigned char) require 16-byte memory alignment:
__attribute__((aligned(16))) float matrix[16];
vec_ld) on PowerPC G4/G5 will silently truncate address bits to 16-byte boundaries or bus error. Use vec_perm or aligned buffers.#ifdef __ALTIVEC__ so non-AltiVec G3 builds compile cleanly.lipo)Every release binary is a fat Mach-O bundle combining PowerPC and Intel/arm64 slices:
lipo -create \
build/bin-ppc750 \
build/bin-ppc7400 \
build/bin-ppc970 \
build/bin-i386 \
build/bin-x86_64 \
build/bin-arm64 \
-output build/game-fat
# Verify slice contents:
lipo -info build/game-fat
GL_ARB_multitextureGL_ARB_vertex_buffer_object (10.4+)GL_EXT_compiled_vertex_arrayglGetString(GL_RENDERER)).