Added run-with-cachegrind script

2025-05-04 10:33:46 +01:00 · 2022-05-23 20:27:17 +03:00 · 2022-05-23 20:27:17 +03:00 · d185f8ad1e
commit d185f8ad1e
parent 7abb18c9ed
1 changed files with 109 additions and 0 deletions
--- a/scripts/run-with-cachegrind.sh
+++ b/scripts/run-with-cachegrind.sh
@ -0,0 +1,109 @@
+#!/bin/bash
+set -euo pipefail
+IFS=$'\n\t'
+
+declare -A event_map
+event_map[Ir]="TotalInstructionsExecuted,executions\n"
+event_map[I1mr]="L1_InstrReadCacheMisses,misses/op\n"
+event_map[ILmr]="LL_InstrReadCacheMisses,misses/op\n"
+event_map[Dr]="TotalMemoryReads,reads\n"
+event_map[D1mr]="L1_DataReadCacheMisses,misses/op\n"
+event_map[DLmr]="LL_DataReadCacheMisses,misses/op\n"
+event_map[Dw]="TotalMemoryWrites,writes\n"
+event_map[D1mw]="L1_DataWriteCacheMisses,misses/op\n"
+event_map[DLmw]="LL_DataWriteCacheMisses,misses/op\n"
+event_map[Bc]="ConditionalBranchesExecuted,executions\n"
+event_map[Bcm]="ConditionalBranchMispredictions,mispredictions/op\n"
+event_map[Bi]="IndirectBranchesExecuted,executions\n"
+event_map[Bim]="IndirectBranchMispredictions,mispredictions/op\n"
+
+now_ms() {
+    echo -n $(date +%s%N | cut -b1-13)
+}
+
+# Run cachegrind on a given benchmark and echo the results.
+CLI_VERSION=$($1 version | tr -d '\n')
+ITERATION_COUNT=$4
+START_TIME=$(now_ms)
+
+valgrind \
+    --quiet \
+    --tool=cachegrind \
+    "$1" run \
+        --load.model model.rbxm \
+        --run "$2" \
+        --headlessRenderer 1 \
+        --lua.globals minSamples=$ITERATION_COUNT \
+        --lua.globals cachegrind=true \
+    >/dev/null
+
+TIME_ELAPSED=$(bc <<< "$(now_ms) - ${START_TIME}")
+
+# Generate report using cg_annotate and extract the header and totals of the
+# recorded events valgrind was configured to record.
+CG_RESULTS=$(cg_annotate $(ls -t cachegrind.out.* | head -1))
+CG_HEADERS=$(grep -B2 'PROGRAM TOTALS$' <<< "$CG_RESULTS" | head -1 | sed -E 's/\s+/\n/g' | sed '/^$/d')
+CG_TOTALS=$(grep 'PROGRAM TOTALS$' <<< "$CG_RESULTS" | head -1 | grep -Po '[0-9,]+\s' | tr -d ', ')
+
+TOTALS_ARRAY=($CG_TOTALS)
+HEADERS_ARRAY=($CG_HEADERS)
+
+declare -A header_map
+for i in "${!TOTALS_ARRAY[@]}"; do
+    header_map[${HEADERS_ARRAY[$i]}]=$i
+done
+
+# Map the results to the format that the benchmark script expects.
+for i in "${!TOTALS_ARRAY[@]}"; do
+    TOTAL=${TOTALS_ARRAY[$i]}
+
+    # Labels and unit descriptions are packed together in the map.
+    EVENT_TUPLE=${event_map[${HEADERS_ARRAY[$i]}]}
+    IFS=$',' read -d '\n' -ra EVENT_VALUES < <(printf "%s" "$EVENT_TUPLE")
+    EVENT_NAME="${EVENT_VALUES[0]}"
+    UNIT="${EVENT_VALUES[1]}"
+
+    case ${HEADERS_ARRAY[$i]} in
+        I1mr | ILmr)
+            REF=${TOTALS_ARRAY[header_map["Ir"]]}
+            OPS_PER_SEC=$(bc -l <<< "$TOTAL / $REF")
+            ;;
+
+        D1mr | DLmr)
+            REF=${TOTALS_ARRAY[header_map["Dr"]]}
+            OPS_PER_SEC=$(bc -l <<< "$TOTAL / $REF")
+            ;;
+
+        D1mw | DLmw)
+            REF=${TOTALS_ARRAY[header_map["Dw"]]}
+            OPS_PER_SEC=$(bc -l <<< "$TOTAL / $REF")
+            ;;
+
+        Bcm)
+            REF=${TOTALS_ARRAY[header_map["Bc"]]}
+            OPS_PER_SEC=$(bc -l <<< "$TOTAL / $REF")
+            ;;
+
+        Bim)
+            REF=${TOTALS_ARRAY[header_map["Bi"]]}
+            OPS_PER_SEC=$(bc -l <<< "$TOTAL / $REF")
+            ;;
+
+        *)
+            OPS_PER_SEC=$(bc -l <<< "$TOTAL")
+            ;;
+        esac
+
+    STD_DEV="0%"
+    RUNS="1"
+
+    if [[ $OPS_PER_SEC =~ ^[+-]?[0-9]*$ ]]
+    then # $OPS_PER_SEC is integer
+        printf "%s#%s x %.0f %s ±%s (%d runs sampled)(roblox-cli version %s)\n" \
+            "$3" "$EVENT_NAME" "$OPS_PER_SEC" "$UNIT" "$STD_DEV" "$RUNS" "$CLI_VERSION"
+    else # $OPS_PER_SEC is float
+        printf "%s#%s x %.10f %s ±%s (%d runs sampled)(roblox-cli version %s)\n" \
+            "$3" "$EVENT_NAME" "$OPS_PER_SEC" "$UNIT" "$STD_DEV" "$RUNS" "$CLI_VERSION"
+    fi
+    
+done