ci: Extract suite splitting to script

Author: posborne

.github/workflows/benchmarks.yml

@@ -50,6 +50,9 @@ jobs:
     strategy:
       matrix:
         partition: ["00", "01", "02", "03", "04", "05", "06", "07"]
+    env:
+      - SPLIT_PREFIX: "benchmarks/split-"
+      - PARTITION_SUITE: "${{ env.SPLIT_PREFIX }}${{ matrix.partition }}.suite"
     steps:
     - uses: actions/checkout@v4
     - name: Downloads All Artifacts (root)
@@ -61,17 +64,10 @@ jobs:
     - run: chmod +x artifacts/*/*.so
     - name: Partition Benchmark Suite
       run: |
-        ceil_div() { echo "($1 + $2 - 1) / $2" | bc; }
-        tmpdir=$(mktemp -d)
-        grep -v '^\#' benchmarks/all.suite > $tmpdir/all.suite
-        line_count=$(wc -l "$tmpdir/all.suite" | awk '{print $1}')
-        lines_per_chunk=$(ceil_div "$line_count" "$PARTITIONS")
-        split -d -l "$lines_per_chunk" "$tmpdir/all.suite" "${tmpdir}/all.split."
-        cp "$tmpdir/all.split.${{ matrix.partition }}" benchmarks/split.${{ matrix.partition }}.suite
-        rm -rf "$tmpdir"
+        benchmarks/split-suite.sh benchmarks/all.suite "${SPLIT_PREFIX}"
 
     - name: Output Running Benchmarks
-      run: cat "benchmarks/split.${{ matrix.partition }}.suite"
+      run: cat "${PARTITION_SUITE}"
 
     - name: Benchmark Partition ${{ matrix.partition }}
       run: |
@@ -80,7 +76,7 @@ jobs:
         --engine=artifacts/wasmtime-v25.0.0/libengine.so \
         --processes=4 \
         --iterations-per-process=1 \
-        -- "benchmarks/split.${{ matrix.partition }}.suite"
+        -- "${PARTITION_SUITE}"
 
   check:
     runs-on: ubuntu-latest

benchmarks/split-suite.sh

@@ -0,0 +1,50 @@
+#!/usr/bin/env sh
+#
+# This script splits a test suite into N roughly
+# equally sized test suites using the provided
+# output prefix.
+#
+# Usage:
+#     ./ci-split.sh <input.suite> <N> <output-prefix>
+#
+# The naming of the split files will being in the form generated
+# by `split` which is `<prefix><num>.suite` where num is in %02d
+# format; for example, a prefix of "benchmarks/split-" will yield
+# files with names like benchmarks/split-07.suite.
+
+set -e
+
+if [ "$#" -ne 3 ]; then
+    echo "Usage: $0 <suite> <number splits> <output prefix>"
+    exit 1
+fi
+
+SUITE="$1"
+NUMBER_SPLITS="$2"
+OUTPUT_PREFIX="$3"
+
+ceil_div() {
+    echo "($1 + $2 - 1) / $2" | bc
+}
+
+count_lines() {
+    wc -l "$1" | awk '{print $1}'
+}
+
+tmpdir="$(mktemp -d)"
+tmp_suite="${tmpdir}/input.suite"
+grep -v '^\#' "${SUITE}" > "${tmp_suite}"
+line_count=$(count_lines "${tmp_suite}")
+lines_per_split=$(ceil_div "$line_count" "$NUMBER_SPLITS")
+split -d -l "$lines_per_split" "${tmp_suite}" "${OUTPUT_PREFIX}"
+rm -rf "$tmpdir"
+
+echo "Generated Splits:"
+for f in "${OUTPUT_PREFIX}"*; do
+    # suite files must end with ".suite" to work correctly
+    # with sigtglass, so do the rename as split isn't
+    # able to do it directly.
+    suited="${f}.suite"
+    mv "$f" "${suited}"
+    echo "- ${suited}"
+done