fix: preserve XMM6-XMM15 in x86_64 bthread context switch

[zchuango]

What problem does this PR solve?

Issue Number: resolve #2781

Problem Summary:

When brpc is built with Clang on Linux x86_64, running
example/streaming_echo_c++ may crash.

This PR preserves XMM6-XMM15 during Linux x86_64 bthread context switches
and updates the corresponding context-frame layout.

What is changed and the side effects?

Changed:

• Save and restore XMM6-XMM15 in bthread_jump_fcontext.
• Expand the x86_64 context frame accordingly.
• Update the MXCSR and x87 control-word offsets.
• Keep bthread_make_fcontext consistent with the new frame layout.
• Update the x86_64 register offsets used by TaskTracer.
• Use unaligned SIMD loads/stores for the saved XMM state.

Side effects:

• Performance effects:

• Breaking backward compatibility:

Verification

Verified on Linux x86_64 with Clang 18.1.3.

After this change:

• streaming_echo_c++ starts and runs normally;
• the client successfully creates the stream;
• the server continuously receives streaming messages;
• the stream is closed normally;
• no core dump is observed.

Environment

Server

Client

---

Check List:

• Please make sure your changes are compilable.
• When providing us with a new feature, it is best to add related tests.
• Please follow Contributor Covenant Code of Conduct.

[Copilot]

Pull request overview

This PR addresses a Linux x86_64 Clang crash (issue #2781) by extending bthread’s x86_64 context-switch frame to preserve SIMD state across bthread_jump_fcontext, and updating dependent tooling that decodes saved contexts.

Changes:

• Save/restore XMM6–XMM15 during Linux x86_64 bthread context switches and expand the saved frame accordingly.
• Update bthread_make_fcontext to match the new frame layout (including updated MXCSR/x87 control-word offsets).
• Update TaskTracer’s register-slot indices to match the new saved-context layout.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/bthread/context.cpp	Expands Linux x86_64 fcontext frame and saves/restores XMM6–XMM15; updates MXCSR/x87 CW offsets and make_fcontext layout.
src/bthread/task_tracer.cpp	Adjusts libunwind cursor register-slot indices to match the expanded Linux x86_64 fcontext frame.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[zchuango]

Hi @wwbmmm @chenBright , the Copilot review comments have been addressed and the updated CI is passing. Could you please take a look when you have time? Thanks!