Bugs fixed in production code: - compare.sh: Python truthiness on 0.0 — `if b_val` was False for 0.0 t/s, displaying it as a dash instead of "0.0". Fixed with `is not None` checks. - compare.sh: ZeroDivisionError when computing delta % with 0.0 baseline. Test improvements (review findings): - detect.bats: kernel param tests now use real detect_kernel_param logic pattern (not a separate reimplementation). Added non-GiB-aligned RAM test, device ID without 0x prefix, empty firmware version, llama-bench detection, detect_total_physical_ram_kb tests. - benchmark_compare.bats: assert delta percentages (+20.0%, -25.0%, 0.0%), test 0.0 t/s edge case, test per-directory error messages, test config change detection with specific field assertions. - log_metrics.bats: add assert_success, --help test, timestamp format validation. Remove unused mock sysfs setup. - common.bats: fix data_dir test, remove redundant assertion, add cleanup. - test_helper.sh: remove unused FIXTURES_DIR. - Remove empty tests/fixtures/ directory. 94 tests, all passing. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
141 lines
4.0 KiB
Bash
141 lines
4.0 KiB
Bash
#!/usr/bin/env bash
|
|
# Compare two benchmark runs side-by-side
|
|
set -euo pipefail
|
|
|
|
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
|
source "$SCRIPT_DIR/../../lib/common.sh"
|
|
source "$SCRIPT_DIR/../../lib/format.sh"
|
|
|
|
if [[ $# -lt 2 ]]; then
|
|
echo "Usage: benchmark compare <before-dir> <after-dir>"
|
|
echo ""
|
|
echo "Examples:"
|
|
echo " bin/benchmark compare data/baselines/20260325-120000 data/benchmarks/post-opt-20260326-100000"
|
|
echo ""
|
|
echo "Available baselines:"
|
|
ls -d "$(data_dir baselines)"/*/ 2>/dev/null | sed 's|/$||' || echo " (none)"
|
|
echo ""
|
|
echo "Available benchmark runs:"
|
|
ls -d "$(data_dir benchmarks)"/*/ 2>/dev/null | sed 's|/$||' || echo " (none)"
|
|
exit 1
|
|
fi
|
|
|
|
BEFORE_DIR="$1"
|
|
AFTER_DIR="$2"
|
|
|
|
for d in "$BEFORE_DIR" "$AFTER_DIR"; do
|
|
if [[ ! -f "$d/summary.json" ]]; then
|
|
log_error "No summary.json in $d"
|
|
exit 1
|
|
fi
|
|
done
|
|
|
|
log_header "Benchmark Comparison"
|
|
|
|
# Extract timestamps from directory names
|
|
before_name="$(basename "$BEFORE_DIR")"
|
|
after_name="$(basename "$AFTER_DIR")"
|
|
log_info "Before: $before_name"
|
|
log_info "After: $after_name"
|
|
|
|
# Show system state diff if available
|
|
if [[ -f "$BEFORE_DIR/system-state.json" ]] && [[ -f "$AFTER_DIR/system-state.json" ]]; then
|
|
echo ""
|
|
python3 - "$BEFORE_DIR/system-state.json" "$AFTER_DIR/system-state.json" << 'PYEOF'
|
|
import sys, json
|
|
|
|
with open(sys.argv[1]) as f:
|
|
before = json.load(f)
|
|
with open(sys.argv[2]) as f:
|
|
after = json.load(f)
|
|
|
|
changes = []
|
|
# Check key config differences
|
|
b_mem = before.get("memory", {})
|
|
a_mem = after.get("memory", {})
|
|
if b_mem.get("vram_total_bytes") != a_mem.get("vram_total_bytes"):
|
|
bv = b_mem.get("vram_total_bytes", 0) / 2**30
|
|
av = a_mem.get("vram_total_bytes", 0) / 2**30
|
|
changes.append(f" VRAM: {bv:.1f} GiB -> {av:.1f} GiB")
|
|
if b_mem.get("gtt_total_bytes") != a_mem.get("gtt_total_bytes"):
|
|
bg = b_mem.get("gtt_total_bytes", 0) / 2**30
|
|
ag = a_mem.get("gtt_total_bytes", 0) / 2**30
|
|
changes.append(f" GTT: {bg:.1f} GiB -> {ag:.1f} GiB")
|
|
|
|
b_kern = before.get("kernel", {})
|
|
a_kern = after.get("kernel", {})
|
|
for param in ["param_iommu", "param_gttsize", "param_pages_limit"]:
|
|
bv = b_kern.get(param, "")
|
|
av = a_kern.get(param, "")
|
|
if bv != av:
|
|
changes.append(f" {param}: '{bv}' -> '{av}'")
|
|
|
|
bt = before.get("tuned_profile", "")
|
|
at = after.get("tuned_profile", "")
|
|
if bt != at:
|
|
changes.append(f" tuned: {bt} -> {at}")
|
|
|
|
if changes:
|
|
print(" Configuration changes:")
|
|
for c in changes:
|
|
print(c)
|
|
else:
|
|
print(" No configuration changes detected")
|
|
PYEOF
|
|
fi
|
|
|
|
# Compare results
|
|
echo ""
|
|
python3 - "$BEFORE_DIR/summary.json" "$AFTER_DIR/summary.json" << 'PYEOF'
|
|
import sys, json
|
|
|
|
with open(sys.argv[1]) as f:
|
|
before = json.load(f)
|
|
with open(sys.argv[2]) as f:
|
|
after = json.load(f)
|
|
|
|
# Index by (model, backend, test)
|
|
def index_results(data):
|
|
idx = {}
|
|
for r in data.get("results", []):
|
|
key = (r["model"], r["backend"], r["test"])
|
|
idx[key] = r["tokens_per_sec"]
|
|
return idx
|
|
|
|
b_idx = index_results(before)
|
|
a_idx = index_results(after)
|
|
|
|
all_keys = sorted(set(b_idx.keys()) | set(a_idx.keys()))
|
|
|
|
if not all_keys:
|
|
print(" No comparable results found.")
|
|
sys.exit(0)
|
|
|
|
fmt = " {:<18} {:<14} {:<7} {:>9} {:>9} {:>8}"
|
|
print(fmt.format("Model", "Backend", "Test", "Before", "After", "Delta"))
|
|
print(" " + "-" * 70)
|
|
|
|
for key in all_keys:
|
|
model, backend, test = key
|
|
b_val = b_idx.get(key)
|
|
a_val = a_idx.get(key)
|
|
|
|
b_str = f"{b_val:.1f}" if b_val is not None else "—"
|
|
a_str = f"{a_val:.1f}" if a_val is not None else "—"
|
|
|
|
if b_val is not None and a_val is not None:
|
|
delta_pct = (a_val - b_val) / b_val * 100 if b_val != 0 else (float('inf') if a_val > 0 else 0)
|
|
if delta_pct > 0:
|
|
d_str = f"\033[32m+{delta_pct:.1f}%\033[0m"
|
|
elif delta_pct < 0:
|
|
d_str = f"\033[31m{delta_pct:.1f}%\033[0m"
|
|
else:
|
|
d_str = "0.0%"
|
|
else:
|
|
d_str = "—"
|
|
|
|
print(fmt.format(model[:18], backend[:14], test, b_str, a_str, d_str))
|
|
|
|
print()
|
|
PYEOF
|