[llvm] [libc] Include CPU model in overlay CI sccache key (PR #196477)

[via llvm-commits]

llvmorg-github-actions[bot] wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-github-workflow

Author: Jeff Bailey (kaladron)

<details>
<summary>Changes</summary>

[libc] Include CPU model in overlay CI sccache key

The overlay CI compiles opt_host memory tests with `-march=native`, which generates object files specific to the runner CPU model. sccache treats `-march=native` as a literal string in its hash key, so cached `.o` files compiled on one CPU model get served to runners with a different CPU. When the cached binary uses instructions the current CPU lacks, the test crashes with SIGILL.

## Symptoms

The `memcmp_opt_host`, `memmove_opt_host`, `memset_opt_host`, `bcmp_opt_host`, and `bzero_opt_host` tests crash immediately when SIMD code paths are exercised (e.g., `Issue77080` for memcmp, `SizeSweep` for memmove/memset). Simple non-SIMD tests pass. The failures are fully reproducible on reruns because the cache stays poisoned.

## Evidence

| Run | Cache Hits | Cache Misses | Result |
|-----|-----------|-------------|--------|
| Clean build (westus3) | 9 | 5354 | PASS |
| Cached build (northcentralus) | 5345 | 0 | CRASH |
| Cached rerun (westus) | 5345 | 0 | CRASH |
| Same-CPU build (eastus) | — | — | PASS |

## Fix

Added a "Detect CPU model" step that reads the CPU model string from `/proc/cpuinfo` (Linux) or `sysctl` (macOS) and appends it to the sccache cache key. Runners with different CPUs now get separate cache buckets.

Assisted-by: Automated tooling, human reviewed.

---
Full diff: https://github.com/llvm/llvm-project/pull/196477.diff


1 Files Affected:

- (modified) .github/workflows/libc-overlay-tests.yml (+20-1) 


``````````diff
diff --git a/.github/workflows/libc-overlay-tests.yml b/.github/workflows/libc-overlay-tests.yml
index 07059e0271af5..96c3e23bc4896 100644
--- a/.github/workflows/libc-overlay-tests.yml
+++ b/.github/workflows/libc-overlay-tests.yml
@@ -44,6 +44,25 @@ jobs:
     - uses: actions/checkout at de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
       with:
         persist-credentials: false
+
+    # The libc build uses -march=native for opt_host tests, which means
+    # the generated object files are specific to the runner's CPU model.
+    # sccache treats -march=native as a literal string in its cache key,
+    # so without per-CPU cache keys, object files compiled on one CPU
+    # model are silently served to runners with a different CPU, causing
+    # illegal instruction crashes at runtime.
+    - name: Detect CPU model
+      id: cpu-info
+      shell: bash
+      run: |
+        if [ "$RUNNER_OS" = "Linux" ]; then
+          cpu_model=$(cat /proc/cpuinfo | grep 'model name' | head -1 | cut -d: -f2 | xargs | tr ' ' '-')
+        elif [ "$RUNNER_OS" = "macOS" ]; then
+          cpu_model=$(sysctl -n machdep.cpu.brand_string | tr ' ' '-')
+        else
+          cpu_model="generic"
+        fi
+        echo "cpu-model=${cpu_model}" >> "$GITHUB_OUTPUT"
     
     # Libc's build is relatively small comparing with other components of LLVM.
     # A fresh linux overlay takes about 180MiB of uncompressed disk space, which can
@@ -56,7 +75,7 @@ jobs:
       uses: hendrikmuhs/ccache-action at 33522472633dbd32578e909b315f5ee43ba878ce # v1.2.22
       with:
         max-size: 1G
-        key: libc_overlay_build_${{ matrix.os }}_${{ matrix.compiler.c_compiler }}
+        key: libc_overlay_build_${{ matrix.os }}_${{ matrix.compiler.c_compiler }}_${{ steps.cpu-info.outputs.cpu-model }}
         variant: sccache
     
     # MPFR is required by some of the mathlib tests.

``````````

</details>


https://github.com/llvm/llvm-project/pull/196477