Adding scripts, commands, and logging scaffolding"

2025-10-04 21:54:21 +00:00
parent 2cc1bb447a
commit f1debadbb3
17 changed files with 292 additions and 0 deletions
--- a/scripts/00_env.sh
+++ b/scripts/00_env.sh
@@ -0,0 +1,21 @@
+#!/bin/bash
+set -eu
+# source this from anywhere inside /home/carlos/projects/gem5
+
+ROOT="/home/carlos/projects/gem5"
+export GEM5="$ROOT/build/ARM/gem5.opt"
+export CFG="$ROOT/scripts/hetero_big_little.py"
+export RUN="$ROOT/gem5-run"
+export OUTROOT="$ROOT/iot/results"
+export LOGROOT="$ROOT/iot/logs"
+
+mkdir -p "$OUTROOT" "$LOGROOT"
+
+# record environment (append-only)
+{
+  echo "==== uname ===="; uname -a
+  echo; echo "==== date ===="; date
+  echo; echo "==== gem5 git ===="; (git -C "$ROOT/gem5src" rev-parse --short HEAD 2>/dev/null || echo n/a)
+} >> "$LOGROOT/env.txt"
+echo "[env] READY"
+
--- a/scripts/10_run_one.sh
+++ b/scripts/10_run_one.sh
@@ -0,0 +1,27 @@
+#!/bin/bash
+set -eu
+source "$(dirname "$0")/00_env.sh"
+
+if [[ $# -lt 5 ]]; then
+  echo "Usage: $0 <workload:{tinyml_kws|sensor_fusion|aes_ccm|attention_kernel}> <core:{big|little|hybrid}> <dvfs:{high|low}> <drowsy:{0|1}> <l2:{512kB|1MB}> [mem=16GB]"
+  exit 1
+fi
+
+W=$1; CORE=$2; DV=$3; DROWSY=$4; L2=$5; MEM=${6:-16GB}
+OUT="$OUTROOT/${W}_${CORE}_${DV}_l2${L2}_d${DROWSY}"
+
+mkdir -p "$OUT"
+echo "[run_one] $W $CORE $DV L2=$L2 drowsy=$DROWSY mem=$MEM -> $OUT"
+
+"$GEM5" "$CFG" \
+  --cmd="$RUN/$W" \
+  --mem="$MEM" \
+  --dvfs="$DV" \
+  --drowsy="$DROWSY" \
+  --l2="$L2" \
+  --outdir="$OUT" \
+  > "$LOGROOT/${W}_${CORE}_${DV}_l2${L2}_d${DROWSY}.stdout.log" \
+  2> "$LOGROOT/${W}_${CORE}_${DV}_l2${L2}_d${DROWSY}.stderr.log"
+
+echo "[run_one] DONE"
+
--- a/scripts/20_sweep.sh
+++ b/scripts/20_sweep.sh
@@ -0,0 +1,23 @@
+#!/bin/bash
+set -eu
+source "$(dirname "$0")/00_env.sh"
+
+run_case () {
+  local W=$1 CORE=$2 DV=$3 D=$4 L2=$5 MEM=16GB
+  bash "$(dirname "$0")/10_run_one.sh" "$W" "$CORE" "$DV" "$D" "$L2" "$MEM"
+}
+
+for W in tinyml_kws sensor_fusion aes_ccm attention_kernel; do
+  for DV in high low; do
+    for D in 0 1; do
+      for L2 in 512kB 1MB; do
+        run_case "$W" big    "$DV" "$D" "$L2"
+        run_case "$W" little "$DV" "$D" "$L2"
+        run_case "$W" hybrid "$DV" "$D" "$L2"
+      done
+    done
+  done
+done
+
+echo "[sweep] ALL DONE"
+
--- a/scripts/30_extract_csv.sh
+++ b/scripts/30_extract_csv.sh
@@ -0,0 +1,28 @@
+#!/bin/bash
+set -eu
+source "$(dirname "$0")/00_env.sh"
+
+CSV="$OUTROOT/phase3_summary.csv"
+echo "workload,core,dvfs,l2,drowsy,sim_seconds,ipc,cycles,insts,l2_miss_rate" > "$CSV"
+
+for d in "$OUTROOT"/*; do
+  [[ -d "$d" ]] || continue
+  base=$(basename "$d")
+  W=$(echo "$base" | cut -d'_' -f1)
+  CORE=$(echo "$base" | cut -d'_' -f2)
+  DVFS=$(echo "$base" | cut -d'_' -f3)
+  L2=$(echo "$base"  | sed -E 's/.*_l2([^_]+).*/\1/')
+  DROW=$(echo "$base" | sed -E 's/.*_d([01]).*/\1/')
+  S="$d/stats.txt"
+
+  SIMS=$(awk '/^sim_seconds/ {print $2}' "$S")
+  IPC=$(awk '/^system\.cpu\.ipc|^system\.cpu0\.ipc/ {print $2}' "$S" | head -n1)
+  CYC=$(awk '/^system\.cpu\.numCycles|^system\.cpu0\.numCycles/ {print $2}' "$S" | head -n1)
+  INST=$(awk '/^system\.cpu\.commit\.committedInsts|^system\.cpu0\.commit\.committedInsts/ {print $2}' "$S" | head -n1)
+  L2MR=$(awk '/^system\.l2\.overall_miss_rate::total/ {print $2}' "$S")
+
+  echo "$W,$CORE,$DVFS,$L2,$DROW,$SIMS,$IPC,$CYC,$INST,$L2MR" >> "$CSV"
+done
+
+echo "[extract] wrote $CSV"
+
--- a/scripts/40_energy_post.py
+++ b/scripts/40_energy_post.py
@@ -0,0 +1,48 @@
+#!/usr/bin/env python3
+import csv, sys, os
+
+root = os.path.dirname(os.path.dirname(__file__))
+src = os.path.join(root, "results", "phase3_summary.csv")
+dst = os.path.join(root, "results", "phase3_summary_energy.csv")
+
+# === your modeling constants (document in Methods) ===
+EPI_PJ = {'big': 200.0, 'little': 80.0, 'hybrid': 104.0}  # pJ/inst
+E_MEM_PJ = 600.0                                          # pJ per L2 miss
+DROWSY_SCALE = 0.85                                       # 15% energy reduction when drowsy=1
+
+rows=[]
+with open(src) as f:
+    r=csv.DictReader(f)
+    for row in r:
+        insts = float(row['insts'])
+        secs  = float(row['sim_seconds'])
+        core  = row['core']
+        drowsy= int(row['drowsy'])
+        epi_pJ= EPI_PJ.get(core, EPI_PJ['little'])
+
+        mr = float(row['l2_miss_rate']) if row['l2_miss_rate'] else 0.0
+        l2_misses = mr * insts  # proxy; replace with MPKI-based calc if available
+
+        energy_instr = (epi_pJ * 1e-12) * insts
+        energy_mem   = (E_MEM_PJ * 1e-12) * l2_misses
+        energy_J     = energy_instr + energy_mem
+        if drowsy == 1:
+            energy_J *= DROWSY_SCALE
+
+        power_W = energy_J / secs if secs > 0 else 0.0
+        edp = energy_J * secs   # CORRECT EDP
+
+        row.update({
+            'energy_J': f"{energy_J:.6f}",
+            'power_W':  f"{power_W:.6f}",
+            'edp':      f"{edp:.6e}",
+            'epi_model_pJ': f"{epi_pJ:.1f}",
+        })
+        rows.append(row)
+
+with open(dst, 'w', newline='') as f:
+    w=csv.DictWriter(f, fieldnames=list(rows[0].keys()))
+    w.writeheader(); w.writerows(rows)
+
+print(f"[energy] wrote {dst}")
+
--- a/scripts/50_plot_epi.py
+++ b/scripts/50_plot_epi.py
@@ -0,0 +1,28 @@
+#!/usr/bin/env python3
+import csv, os
+import matplotlib.pyplot as plt
+from collections import defaultdict
+
+root = os.path.dirname(os.path.dirname(__file__))
+src = os.path.join(root, "results", "phase3_summary_energy.csv")
+out = os.path.join(root, "results", "fig_epi_across_workloads.png")
+
+epi_by_core = defaultdict(list)
+with open(src) as f:
+    r=csv.DictReader(f)
+    for row in r:
+        insts=float(row['insts']); energy=float(row['energy_J'])
+        epi = 1e12*energy/insts if insts>0 else 0.0
+        epi_by_core[row['core']].append(epi)
+
+cores=['big','little','hybrid']
+vals=[sum(epi_by_core[c])/max(1,len(epi_by_core[c])) for c in cores]
+
+plt.figure()
+plt.bar(cores, vals)
+plt.ylabel('EPI (pJ/inst)')
+plt.title('Energy per Instruction across Workloads (avg by core mode)')
+plt.tight_layout()
+plt.savefig(out)
+print(f"[plot] wrote {out}")
+
--- a/scripts/51_plot_edp_tinyml.py
+++ b/scripts/51_plot_edp_tinyml.py
@@ -0,0 +1,25 @@
+#!/usr/bin/env python3
+import csv, os
+import matplotlib.pyplot as plt
+
+root = os.path.dirname(os.path.dirname(__file__))
+src = os.path.join(root, "results", "phase3_summary_energy.csv")
+out = os.path.join(root, "results", "fig_tinyml_edp.png")
+
+labels=[]; edps=[]
+with open(src) as f:
+    r=csv.DictReader(f)
+    for row in r:
+        if row['workload']!='tinyml_kws': continue
+        labels.append(f"{row['core']}-{row['dvfs']}-L2{row['l2']}-d{row['drowsy']}")
+        edps.append(float(row['edp']))
+
+plt.figure()
+plt.bar(labels, edps)
+plt.ylabel('EDP (J·s)')
+plt.title('TinyML: EDP by configuration')
+plt.xticks(rotation=60, ha='right')
+plt.tight_layout()
+plt.savefig(out)
+print(f"[plot] wrote {out}")
+
--- a/scripts/60_bundle_logs.sh
+++ b/scripts/60_bundle_logs.sh
@@ -0,0 +1,23 @@
+#!/bin/bash
+set -eu
+source "$(dirname "$0")/00_env.sh"
+
+# terminal excerpts
+: > "$LOGROOT/TERMINAL_EXCERPTS.txt"
+for f in "$LOGROOT"/*.stdout.log; do
+  echo "===== $(basename "$f") =====" >> "$LOGROOT/TERMINAL_EXCERPTS.txt"
+  (head -n 20 "$f"; echo "..."; tail -n 20 "$f") >> "$LOGROOT/TERMINAL_EXCERPTS.txt"
+  echo >> "$LOGROOT/TERMINAL_EXCERPTS.txt"
+done
+echo "[bundle] wrote $LOGROOT/TERMINAL_EXCERPTS.txt"
+
+# stats excerpts
+: > "$LOGROOT/STATS_EXCERPTS.txt"
+for d in "$OUTROOT"/*; do
+  [[ -d "$d" ]] || continue
+  echo "===== $(basename "$d") =====" >> "$LOGROOT/STATS_EXCERPTS.txt"
+  awk '/^sim_seconds|^system\.cpu\.ipc|^system\.cpu0\.ipc|^system\.cpu\.numCycles|^system\.cpu0\.numCycles|^system\.cpu\.commit\.committedInsts|^system\.cpu0\.commit\.committedInsts|^system\.l2\.overall_miss_rate::total/' "$d/stats.txt" >> "$LOGROOT/STATS_EXCERPTS.txt"
+  echo >> "$LOGROOT/STATS_EXCERPTS.txt"
+done
+echo "[bundle] wrote $LOGROOT/STATS_EXCERPTS.txt"
+
--- a/scripts/70_diff_table.py
+++ b/scripts/70_diff_table.py
@@ -0,0 +1,34 @@
+#!/usr/bin/env python3
+import csv, os
+root = os.path.dirname(os.path.dirname(__file__))
+src = os.path.join(root, "results", "phase3_summary_energy.csv")
+dst = os.path.join(root, "results", "phase3_drowsy_deltas.csv")
+
+# group by key without drowsy; compare d0 vs d1
+from collections import defaultdict
+bykey = defaultdict(dict)
+
+with open(src) as f:
+    r=csv.DictReader(f)
+    for row in r:
+        key = (row['workload'], row['core'], row['dvfs'], row['l2'])
+        bykey[key][row['drowsy']] = row
+
+rows=[]
+for k, d in bykey.items():
+    if '0' in d and '1' in d:
+        a=d['0']; b=d['1']
+        e0=float(a['energy_J']); e1=float(b['energy_J'])
+        edp0=float(a['edp']);    edp1=float(b['edp'])
+        rows.append({
+            'workload':k[0],'core':k[1],'dvfs':k[2],'l2':k[3],
+            'energy_drop_%': f"{100*(e0-e1)/e0:.2f}",
+            'edp_drop_%':    f"{100*(edp0-edp1)/edp0:.2f}"
+        })
+
+with open(dst,'w',newline='') as f:
+    w=csv.DictWriter(f, fieldnames=list(rows[0].keys()))
+    w.writeheader(); w.writerows(rows)
+
+print(f"[delta] wrote {dst}")
+