feat: accelerate CI build speed by using pre-built pulls.

[Cyberhan123]

Summary

This pull request adds support for using prebuilt GGML SDKs as an alternative to building GGML from source or using a system-installed version. It introduces new CMake options and logic to locate, download, extract, and integrate a prebuilt GGML package, improving build flexibility and simplifying downstream packaging. Documentation and contribution guidelines are updated to reflect this new workflow and its requirements.

Support for prebuilt GGML SDKs:

• Added the SD_USE_PREBUILT_GGML CMake option and related variables (e.g., GGML_ROOT_DIR, GGML_RELEASE_BASE_URL, GGML_RELEASE_TAG, etc.) to allow specifying or downloading a prebuilt GGML SDK instead of building from source or using a system library. Mutual exclusion with SD_USE_SYSTEM_GGML is enforced. (CMakeLists.txt)
• Integrated new logic in CMakeLists.txt to handle prebuilt GGML SDKs: imports the SDK, sets up CMake targets, and ensures runtime binaries are copied if needed. (CMakeLists.txt) [1] [2]
• Added the cmake/fetch_prebuilt_ggml.cmake module, which implements platform detection, archive download and extraction, package root location, and CMake target patching for prebuilt GGML SDKs.

Documentation and contribution guidelines:

• Updated docs/build.md with instructions for building using a prebuilt GGML SDK, including both unpacked and downloadable workflows, and notes on compatibility requirements.
• Clarified in CONTRIBUTING.md that prebuilt GGML SDKs must match the validated GGML revision and ABI-sensitive options (such as GGML_MAX_NAME=128).

technologies

To add some information about the related technologies:
https://github.com/seasonjs/ggml.cpp-build provides a pre-build option, allowing for the direct construction of a unified cross-platform package. While this increases the download size, it's possible to parse the manifests.json file and then split it into CUDA, HIP, and CPU components. However, this would require writing a script for parsing, and I don't want this PR to be too large. Therefore, I think this feature can be implemented later.
The successfully implemented prototype is available at https://github.com/seasonjs/stable-diffusion.cpp-build.

performance

Speed ​​and performance have seen an exponential improvement, reducing the time from one hour to just four minutes.

Checklist

• I have read and confirmed this PR follows the contribution guidelines.

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

This PR adds first-class support for consuming a prebuilt ggml SDK (either unpacked locally or downloaded from a release) to speed up builds/packaging, and updates CI packaging jobs to use that prebuilt ggml flow.

Changes:

• Introduces SD_USE_PREBUILT_GGML and related GGML_* CMake cache variables to import ggml from an SDK instead of building the bundled submodule.
• Adds a new CMake helper (cmake/fetch_prebuilt_ggml.cmake) to download/cache/extract and patch imported ggml targets, plus runtime-binary copying.
• Updates CI build matrix and artifact packaging to build against the prebuilt ggml SDK, and documents the workflow/ABI constraints.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
docs/build.md	Documents how to build using an unpacked or downloadable prebuilt ggml SDK.
cmake/fetch_prebuilt_ggml.cmake	Implements downloading/caching/extracting and importing of prebuilt ggml SDKs + runtime copy helper.
CONTRIBUTING.md	Clarifies ABI/build-option requirements for prebuilt ggml SDKs.
CMakeLists.txt	Adds prebuilt-ggml option & wires it into dependency setup and runtime copying.
.github/workflows/build.yml	Switches CI builds/artifacts to use prebuilt ggml SDK flow and simplifies build jobs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.