#14406: Fix BW computation

tenstorrent · Nov 8, 2024 · fefe768 · fefe768
1 parent 3ba4c54
commit fefe768
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 5 deletions.
diff --git a/tests/ttnn/unit_tests/operations/ccl/perf/perf_csv.py b/tests/ttnn/unit_tests/operations/ccl/perf/perf_csv.py
@@ -12,7 +12,7 @@ def perf_report(file_path):
 
     df = df.dropna(subset=["DEVICE ERISC KERNEL DURATION [ns]"])
     df = df[df["OP TO OP LATENCY [ns]"] != 0]
-    df = df[df["TRACE ID"].notna() & (df["TRACE ID"] != "")]
+    df = df[df["METAL TRACE ID"].notna() & (df["METAL TRACE ID"] != "")]
 
     def remove_keys_from_attributes(attributes):
         attributes = attributes.replace(";", ",").replace("'", '"')
@@ -156,15 +156,15 @@ def calculate_bandwidth(row):
                 op_bw = (output_tensor_volume * (n_chips - 1) / n_chips) / longest_device_fw_time
                 link_bw = (output_tensor_volume * (n_chips - 1) / n_chips) / longest_erisc_fw_time
             elif row["OP CODE"] == "ReduceScatter":
-                op_bw = (input_tensor_volume / n_chips) / longest_device_fw_time
-                link_bw = (input_tensor_volume * (n_chips - 1) / n_chips) / longest_erisc_fw_time
+                op_bw = input_tensor_volume / longest_device_fw_time
+                link_bw = input_tensor_volume / longest_erisc_fw_time
         elif row["topology"] == "Linear":
             if row["OP CODE"] == "AllGather":
                 op_bw = input_tensor_volume * n_chips / longest_device_fw_time
                 link_bw = input_tensor_volume * (n_chips - 1) / longest_erisc_fw_time
             elif row["OP CODE"] == "ReduceScatter":
                 op_bw = input_tensor_volume / longest_device_fw_time
-                link_bw = input_tensor_volume * (n_chips - 1) / n_chips / longest_erisc_fw_time
+                link_bw = input_tensor_volume / longest_erisc_fw_time
         return round(op_bw, 2), round(link_bw, 2)
 
     for i, (group, group_df) in enumerate(grouped, start=1):
@@ -196,6 +196,7 @@ def calculate_bandwidth(row):
             "output_mem_config": group_df["output_mem_config"].iloc[0] if "output_mem_config" in group_df else "",
             "topology": group_df["topology"].iloc[0],
             "Layout": group_df["Layout"].iloc[0] if "Layout" in group_df else "",
+            "Data Type": group_df["Data Type"].iloc[0] if "Data Type" in group_df else "",
         }
 
         for column in numeric_columns:

diff --git a/tests/ttnn/unit_tests/operations/ccl/perf/test_ccl_perf.py b/tests/ttnn/unit_tests/operations/ccl/perf/test_ccl_perf.py
@@ -154,7 +154,6 @@ def test_all_gather_on_t3000(
     "input_dtype",
     [
         ttnn.bfloat16,
-        # ttnn.bfloat8_b,
     ],
 )
 @pytest.mark.parametrize(