Added optional HBM results

.github/workflows/triton-benchmarks.yml

@@ -113,7 +113,9 @@ jobs:
           source /home/runner/intel/oneapi/mkl/2024.1/env/vars.sh
           cd benchmarks/xetla_benchmark
           python fused_softmax.py --reports $REPORTS
-          python build_report.py $REPORTS/softmax-performance.csv $REPORTS/softmax-report.csv --param_cols "N" --result_cols "Triton-TFlops-max,XeTLA-TFlops-max" --name softmax
+          source ../../scripts/capture-hw-details.sh
+          python ../../scripts/build_report.py $REPORTS/softmax-performance.csv $REPORTS/softmax-report.csv --benchmark softmax --compiler triton --param_cols "N" --tflops_col Triton-TFlops-max --hbm_col "Triton-GB/s-max"
+          python ../../scripts/build_report.py $REPORTS/softmax-performance.csv $REPORTS/softmax-report.csv --benchmark softmax --compiler xetla --param_cols "N" --tflops_col XeTLA-TFlops-max --hbm_col "XeTLA-GB/s-max"
 
 
       - name: Run micro benchmark

scripts/build_report.py

@@ -9,15 +9,16 @@
 
 def parse_args():
     parser = argparse.ArgumentParser(description="Build report based on triton-benchmark run")
-    parser.add_argument("source", help="Path to source csv file with benchmark results", required=True)
+    parser.add_argument("source", help="Path to source csv file with benchmark results")
     parser.add_argument(
         "target",
-        required=True,
         help="Path to result csv file with benchmark results including host info and dates",
     )
     parser.add_argument("--param_cols", help="Names of parameter columns, separated by commas.", required=True)
-    parser.add_argument("--name", help="Name of the benchmark.", required=True)
-    parser.add_argument("--result_cols", help="Names of the result columns, separated by commas.", required=True)
+    parser.add_argument("--benchmark", help="Name of the benchmark.", required=True)
+    parser.add_argument("--compiler", help="Name of the compiler, like `triton`.", required=True)
+    parser.add_argument("--tflops_col", help="Column name with tflops.", required=True)
+    parser.add_argument("--hbm_col", help="Column name with HBM results.", required=False, default=None)
     return parser.parse_args()
 
 
@@ -26,54 +27,40 @@ def check_cols(target_cols, all_cols):
     assert (len(diff) == 0), f"Couldn't find required columns: '{diff}' among available '{all_cols}'"
 
 
-def parse_result(name):
-    name = name.lower()
-    if "triton" in name:
-        return "triton"
-    elif "xetla" in name:
-        return "xetla"
-    else:
-        return name
-
-
-def transform_df(df, param_cols, result_cols, bench_name):
-
+def transform_df(df, param_cols, tflops_col, hbm_col, benchmark, compiler):
     check_cols(param_cols, df.columns)
-    check_cols(result_cols, df.columns)
+    check_cols([tflops_col] + [] if hbm_col is None else [hbm_col], df.columns)
     # Build json with parameters
-    df_results = df[result_cols].copy()
+    df_results = pd.DataFrame()
     df_results["params"] = [json.dumps(j) for j in df[param_cols].astype(int).to_dict("records")]
+    df_results['tflops']  = df[tflops_col]
+    if hbm_col is not None:
+        df_results['hbm_gbs'] = df[hbm_col]
 
-    df = pd.melt(
-        df_results,
-        id_vars=["params"],
-        value_vars=result_cols,
-        value_name="tflops",
-        var_name="comment",
-    )
-
-    df["compiler"] = df["comment"].apply(parse_result)
-    df["run_uuid"] = uuid.uuid4().hex
-    df["datetime"] = datetime.datetime.now()
-    df["benchmark"] = bench_name
+    df_results["run_uuid"] = uuid.uuid4().hex
+    df_results["datetime"] = datetime.datetime.now()
+    df_results["benchmark"] = benchmark
+    df_results["compiler"] = compiler
 
     host_info = {
         n: os.getenv(n.upper(), default="")
         for n in ["libigc1_version", "level_zero_version", "gpu_device", "agama_version"]
     }
     assert host_info['gpu_device'], "Could not find GPU device description, was capture_device.sh called?"
     for name, val in host_info.items():
-        df[name] = val
+        df_results[name] = val
 
-    return df
+    return df_results
 
 
 def main():
     args = parse_args()
     param_cols = args.param_cols.split(",")
-    result_cols = args.result_cols.split(",")
     df = pd.read_csv(args.source)
-    result_df = transform_df(df, param_cols=param_cols, result_cols=result_cols, bench_name=args.name)
+    result_df = transform_df(
+        df, param_cols=param_cols, tflops_col=args.tflops_col,
+        hbm_col=args.hbm_col, benchmark=args.benchmark, compiler=args.compiler
+    )
     result_df.to_csv(args.target, index=False)