ECE 408 sem project

verify_nsys.slurm

@@ -0,0 +1,49 @@
+#!/bin/bash
+#SBATCH --job-name="verify"
+#SBATCH --partition=gpuA40x4
+#SBATCH --mem=12G
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=1
+#SBATCH --cpus-per-task=1
+#SBATCH --constraint="projects,perf,nvperf"
+#SBATCH --gpus-per-node=1
+#SBATCH --gpu-bind=closest
+#SBATCH --account=bche-delta-gpu
+#SBATCH -t 00:15:00
+#SBATCH --exclude=gpub029,gpub034,gpub038,gpub043,gpub074,gpub054,gpub012
+
+module reset
+module load cuda
+
+SUFFIX=${1:-op9}
+BIN=./build/test_gpt2_${SUFFIX}
+
+# -------------------------
+# output dirs
+# -------------------------
+OUTDIR=output/perf/nsys
+REPORTDIR=output/perf/reports
+
+mkdir -p $OUTDIR
+mkdir -p $REPORTDIR
+
+OUTFILE=$OUTDIR/verify_${SUFFIX}.out
+ERRFILE=$OUTDIR/verify_${SUFFIX}.err
+
+REPORT=$REPORTDIR/report_${SUFFIX}
+
+echo "Binary: $BIN"
+echo "NSYS report: $REPORT"
+
+# redirect only SLURM logs
+exec > $OUTFILE 2> $ERRFILE
+
+echo -e "job $SLURM_JOBID starting on $(hostname)\n\n"
+
+# -------------------------
+# NSYS profiling
+# -------------------------
+srun nsys profile \
+    --stats=true \
+    --output=$REPORT \
+    $BIN