[CI] add LLVM IR bloat regression check

.github/scripts/compare-llvm-lines.sh

@@ -0,0 +1,57 @@
+#!/usr/bin/env bash
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT license.
+#
+# Compare two cargo-llvm-lines output files and produce a markdown regression report.
+#
+# Usage:
+#   ./compare-llvm-lines.sh <baseline.txt> <current.txt>
+
+set -euo pipefail
+
+if [ $# -lt 2 ]; then
+  echo "Usage: $0 <baseline.txt> <current.txt>" >&2
+  exit 1
+fi
+
+baseline_file="$1"
+current_file="$2"
+
+baseline_total=$(awk '/\(TOTAL\)/{print $1}' "$baseline_file")
+current_total=$(awk '/\(TOTAL\)/{print $1}' "$current_file")
+
+if [ "$baseline_total" -eq 0 ]; then
+  echo "Error: baseline total is 0, cannot compute growth." >&2
+  exit 1
+fi
+
+delta_total=$(( current_total - baseline_total ))
+
+# Parse llvm-lines output into "lines\tfunction_name" format.
+# Skip header (3 lines) and TOTAL row. Extract the first number (lines count)
+# and the function name (everything after the second parenthesized group).
+parse() {
+  awk 'NR>3 && !/TOTAL/{
+    lines = $1
+    sub(/^[^)]*\)[^)]*\) */, "")
+    print lines "\t" $0
+  }' "$1" | sort -t$'\t' -k2
+}
+
+tmpdir=$(mktemp -d)
+trap 'rm -rf "$tmpdir"' EXIT
+baseline_parsed_file="$tmpdir/llvm_baseline_parsed.txt"
+current_parsed_file="$tmpdir/llvm_current_parsed.txt"
+
+parse "$baseline_file" > "$baseline_parsed_file"
+parse "$current_file" > "$current_parsed_file"
+
+echo "| Delta | Current | Baseline | Function |"
+echo "|-------|---------|----------|----------|"
+printf "| %+d | %s | %s | (TOTAL) |\n" "$delta_total" "$current_total" "$baseline_total"
+join -t$'\t' -j 2 -a1 -a2 -e 0 -o 0,1.1,2.1 \
+  "$current_parsed_file" "$baseline_parsed_file" | \
+  awk -F'\t' '{delta=$2-$3; printf "%d\t%s\t%s\t%s\n", delta, $2, $3, $1}' | \
+  sort -t$'\t' -k1,1rn -k2,2rn | \
+  awk -F'\t' '{printf "| %+d | %s | %s | %s |\n", $1, $2, $3, $4}'
+echo ""

.github/workflows/ci.yml

@@ -487,3 +487,59 @@ jobs:
         env:
           CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
 
+  # LLVM IR bloat check: compare monomorphization cost against baseline.
+  # This job is not part of the required CI gate, so it won't block PR merges.
+  llvm-lines:
+    if: github.event_name == 'pull_request'
+    needs: basics
+    name: LLVM Lines Regression Check
+    runs-on: ubuntu-latest
+    env:
+      LLVM_LINES_GROWTH_THRESHOLD: 5
+    steps:
+      - name: Checkout current branch
+        uses: actions/checkout@v4
+        with:
+          path: pr
+
+      - name: Checkout base ref
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.base.sha }}
+          path: baseline
+
+      - name: Install Rust
+        shell: bash
+        run: rustup show
+        working-directory: pr
+
+      - name: Install cargo-llvm-lines
+        run: cargo install cargo-llvm-lines --locked
+
+      - uses: Swatinem/rust-cache@v2
+        with:
+          workspaces: |
+            pr -> target
+            baseline -> target
+
+      - name: Generate baseline LLVM lines
+        working-directory: baseline
+        run: cargo llvm-lines --package diskann-benchmark --all-features --release | head -100 | tee ../baseline-llvm-lines.txt
+
+      - name: Generate current LLVM lines
+        working-directory: pr
+        run: cargo llvm-lines --package diskann-benchmark --all-features --release | head -100 | tee ../current-llvm-lines.txt
+
+      - name: Compare LLVM lines
+        run: bash pr/.github/scripts/compare-llvm-lines.sh baseline-llvm-lines.txt current-llvm-lines.txt
+
+      - name: Check LLVM lines growth threshold
+        run: |
+          baseline_total=$(awk '/\(TOTAL\)/{print $1}' baseline-llvm-lines.txt)
+          current_total=$(awk '/\(TOTAL\)/{print $1}' current-llvm-lines.txt)
+          growth=$(( (current_total - baseline_total) * 100 / baseline_total ))
+
+          if [ "$growth" -gt "$LLVM_LINES_GROWTH_THRESHOLD" ]; then
+            echo "::error::LLVM IR grew ${growth}% ($baseline_total → $current_total lines), threshold is ${LLVM_LINES_GROWTH_THRESHOLD}%"
+            exit 1
+          fi