cmake : skip project() when consumed as a subdirectory (#20415)

[jinweihan-ai]

Summary

Fixes #20415.

Both CMakeLists.txt (llama) and ggml/CMakeLists.txt call project() unconditionally. When llama.cpp is pulled into another CMake project via add_subdirectory / FetchContent / CPM, those calls re-run toolchain detection inside the nested scope. For C/CXX this is mostly a no-op, but ggml's project(... ASM) triggers an ASM compiler identification pass that can overwrite parent-scope toolchain variables and, on cross-compilation toolchains (iOS, Android NDK, arm64↔x86_64 hosts), pick a different sysroot/architecture than the parent configured.

The fix follows the CMake convention of guarding project() with the standalone check so it only fires for top-level builds. The existing LLAMA_STANDALONE / GGML_STANDALONE variables continue to work because they are computed with the same predicate. No assembly sources live at the top-level of ggml — all ASM usage is gated behind enable_language(ASM) inside the backend that needs it (Metal, Hexagon HTP), so dropping ASM from the embedded project() call is safe.

Repro (before the fix)

Minimal parent project:

cmake_minimum_required(VERSION 3.20)
project(parent_app C CXX)                # parent enables C/CXX only, no ASM
add_subdirectory(/path/to/llama.cpp llama_build)

$ cmake -B build -DCMAKE_OSX_ARCHITECTURES=x86_64 ...
-- The C compiler identification is AppleClang
-- The CXX compiler identification is AppleClang
...
-- The ASM compiler identification is AppleClang       ← re-triggered by ggml's project()
-- Found assembler: /usr/bin/cc                         ← leaks into parent scope

After the fix those two lines no longer appear — the parent's toolchain state is preserved.

Test plan

• Standalone llama.cpp build still configures and compiles end-to-end (cmake -B build && cmake --build build --target llama-cli), version reported correctly (built with AppleClang … for Darwin arm64).
• Minimal parent project with add_subdirectory(llama.cpp) + -DCMAKE_OSX_ARCHITECTURES=x86_64 now configures without triggering ASM identification and without clobbering CMAKE_ASM_COMPILER in the parent scope.
• GGML_SYSTEM_ARCH is still detected correctly (x86 in the cross-compile repro).
• CI ggml-ci, Android build, and standalone builds should remain green — only the project() invocation is guarded; everything below it runs unchanged.

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: Yes. Patch drafting and PR description were AI-assisted; the subdirectory repro was built and the standalone build verified locally before submission.

[ggml-gh-bot]

Hi @jinweihan-ai, thanks for your contribution!

Per our contribution guidelines, the automated PR checker found the following issue(s) that need your attention:

• Multiple open PRs from a new contributor: We limit new contributors (those without a previously merged PR) to 1 open PR at a time. You currently have 2 open PRs.

• AI-generated content: This project does not accept PRs, descriptions or commit messages that are fully or predominantly AI-generated. If you have used AI to assist you in writing code, please make sure to disclose that explicitly.

---

Please note that maintainers reserve the right to make final decisions on PRs. If you believe there is a mistake, please comment below.