Build for testing baseline builds

[dylan-conway]

No description provided.

[coderabbitai]

Walkthrough

Refactored the JIT probe trampoline mechanism from a conditional macro-based approach with AVX variants to a unified inline assembly implementation. Removed the ctiMasmProbeTrampolineAVX declaration and consolidated the probe setup, stack handling, and state restoration logic into a single trampoline path.

Changes

Cohort / File(s)	Summary
Probe Trampoline Refactoring Source/JavaScriptCore/assembler/MacroAssemblerX86_64.cpp	Removed AVX-specific trampoline declaration (ctiMasmProbeTrampolineAVX) and replaced conditional macro-based trampoline selection with a single inline asm sequence. Consolidated stack frame management, parameter saving, probe invocation, and state restoration into unified implementation. Net change: +191/-265 lines.

🚥 Pre-merge checks | ❌ 2

❌ Failed checks (1 warning, 1 inconclusive)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	No pull request description was provided; the template requires a bug reference, explanation, and file change details, all of which are missing.	Add a description following the WebKit template with: a Bugzilla bug reference, explanation of the changes, reviewer attribution, and list of modified files with function/class details.
Title check	❓ Inconclusive	The title 'Build for testing baseline builds' is vague and generic, using non-descriptive terms that don't clearly convey what actual changes were made to the codebase.	Replace with a more specific title that describes the main technical change, such as 'Remove AVX-specific trampoline and consolidate probe invocation to single path' or 'Refactor MacroAssemblerX86_64 probe trampoline implementation'.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🤖 Fix all issues with AI agents

In `@Source/JavaScriptCore/assembler/MacroAssemblerX86_64.cpp`:
- Around line 220-235: The trampoline currently uses AVX-only vmovaps
instructions to save/restore XMM registers (the sequence writing to
PROBE_CPU_XMM0...PROBE_CPU_XMM15_VECTOR_OFFSET), which will fault on non-AVX
CPUs; either add a runtime AVX gate back into the trampoline or change the
save/restore sequence to SSE instructions (use movaps instead of vmovaps for the
lines that reference PROBE_CPU_XMM*_VECTOR_OFFSET) so the trampoline is safe on
all x86_64 targets; update both the save block (the shown vmovaps sequence) and
the corresponding restore block (the similar sequence at lines ~297-312)
consistently.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Avoid unconditional AVX-only instructions in the trampoline.

vmovaps requires AVX. With the AVX/non‑AVX dispatch removed, running probes on non‑AVX x86_64 will trap. Either reintroduce the runtime AVX gating or switch these to SSE movaps so the trampoline is safe on all x86_64 targets.

🛠️ Suggested SSE fallback (use movaps for all XMM saves/restores)

-    "vmovaps %xmm0,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM0_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm1,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM1_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm2,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM2_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm3,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM3_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm4,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM4_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm5,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM5_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm6,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM6_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm7,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM7_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm8,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM8_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm9,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM9_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm10, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM10_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm11, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM11_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm12, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM12_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm13, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM13_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm14, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM14_VECTOR_OFFSET) "(%rsp)" "\n"
-    "vmovaps %xmm15, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM15_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm0,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM0_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm1,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM1_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm2,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM2_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm3,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM3_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm4,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM4_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm5,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM5_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm6,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM6_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm7,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM7_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm8,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM8_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm9,  " STRINGIZE_VALUE_OF(PROBE_CPU_XMM9_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm10, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM10_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm11, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM11_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm12, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM12_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm13, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM13_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm14, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM14_VECTOR_OFFSET) "(%rsp)" "\n"
+    "movaps %xmm15, " STRINGIZE_VALUE_OF(PROBE_CPU_XMM15_VECTOR_OFFSET) "(%rsp)" "\n"
...
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM0_VECTOR_OFFSET) "(%rbp), %xmm0" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM1_VECTOR_OFFSET) "(%rbp), %xmm1" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM2_VECTOR_OFFSET) "(%rbp), %xmm2" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM3_VECTOR_OFFSET) "(%rbp), %xmm3" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM4_VECTOR_OFFSET) "(%rbp), %xmm4" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM5_VECTOR_OFFSET) "(%rbp), %xmm5" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM6_VECTOR_OFFSET) "(%rbp), %xmm6" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM7_VECTOR_OFFSET) "(%rbp), %xmm7" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM8_VECTOR_OFFSET) "(%rbp), %xmm8" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM9_VECTOR_OFFSET) "(%rbp), %xmm9" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM10_VECTOR_OFFSET) "(%rbp), %xmm10" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM11_VECTOR_OFFSET) "(%rbp), %xmm11" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM12_VECTOR_OFFSET) "(%rbp), %xmm12" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM13_VECTOR_OFFSET) "(%rbp), %xmm13" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM14_VECTOR_OFFSET) "(%rbp), %xmm14" "\n"
-    "vmovaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM15_VECTOR_OFFSET) "(%rbp), %xmm15" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM0_VECTOR_OFFSET) "(%rbp), %xmm0" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM1_VECTOR_OFFSET) "(%rbp), %xmm1" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM2_VECTOR_OFFSET) "(%rbp), %xmm2" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM3_VECTOR_OFFSET) "(%rbp), %xmm3" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM4_VECTOR_OFFSET) "(%rbp), %xmm4" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM5_VECTOR_OFFSET) "(%rbp), %xmm5" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM6_VECTOR_OFFSET) "(%rbp), %xmm6" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM7_VECTOR_OFFSET) "(%rbp), %xmm7" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM8_VECTOR_OFFSET) "(%rbp), %xmm8" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM9_VECTOR_OFFSET) "(%rbp), %xmm9" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM10_VECTOR_OFFSET) "(%rbp), %xmm10" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM11_VECTOR_OFFSET) "(%rbp), %xmm11" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM12_VECTOR_OFFSET) "(%rbp), %xmm12" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM13_VECTOR_OFFSET) "(%rbp), %xmm13" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM14_VECTOR_OFFSET) "(%rbp), %xmm14" "\n"
+    "movaps " STRINGIZE_VALUE_OF(PROBE_CPU_XMM15_VECTOR_OFFSET) "(%rbp), %xmm15" "\n"

Also applies to: 297-312

🤖 Prompt for AI Agents

In `@Source/JavaScriptCore/assembler/MacroAssemblerX86_64.cpp` around lines 220 -
235, The trampoline currently uses AVX-only vmovaps instructions to save/restore
XMM registers (the sequence writing to
PROBE_CPU_XMM0...PROBE_CPU_XMM15_VECTOR_OFFSET), which will fault on non-AVX
CPUs; either add a runtime AVX gate back into the trampoline or change the
save/restore sequence to SSE instructions (use movaps instead of vmovaps for the
lines that reference PROBE_CPU_XMM*_VECTOR_OFFSET) so the trampoline is safe on
all x86_64 targets; update both the save block (the shown vmovaps sequence) and
the corresponding restore block (the similar sequence at lines ~297-312)
consistently.

[github-actions]

Preview Builds

Commit	Release	Date
8fb3f004	autobuild-preview-pr-151-8fb3f004	2026-01-30 03:17:40 UTC