[xcl-lens] Support pcie mem channel

projects/xcl-lens/src/xcl_lens/parser/rccl.py

@@ -31,6 +31,7 @@ class RcclLogParser:
         self._report_channel_transport_info()
         self._report_collective_transfers()
         self._report_p2p_transfers()
+        self._report_pcie_mem_channel_info()
 
         print(" End of Report ".center(80, "="))
 
@@ -206,6 +207,9 @@ class RcclLogParser:
                    - pattern: Regex pattern to match the field key (e.g., r"protocol")
                    - col_name: Name of the DataFrame column
                    - value_pattern: Regex pattern to validate/extract the field value
+                   A 3-item tuple (col_name, value_pattern, literal_value) is also
+                   supported for literal/default columns. If value_pattern is None,
+                   literal_value is assigned directly.
             mandatory: List of column names that must not be NaN (drop rows missing these)
             verbose_cols: List of column names to keep when not verbose
             sort_cols: List of column names to sort by (in order)
@@ -221,11 +225,29 @@ class RcclLogParser:
 
         # Create DataFrame and extract all fields using regex with validation
         df = pd.DataFrame(data, columns=["host", "rank", "raw_log"])
-        for pattern, (col_name, val_pattern) in fields.items():
-            # Extract field with strict value validation using word boundary
-            df[col_name] = df["raw_log"].str.extract(
-                rf"\b{pattern}\s+({val_pattern})", expand=False
-            )
+        for pattern, field_spec in fields.items():
+            if len(field_spec) == 2:
+                col_name, val_pattern = field_spec
+                # Extract field with strict value validation using word boundary
+                df[col_name] = df["raw_log"].str.extract(
+                    rf"\b{pattern}\s+({val_pattern})", expand=False
+                )
+            elif len(field_spec) == 3:
+                col_name, val_pattern, literal_value = field_spec
+                if val_pattern is None:
+                    matched = df["raw_log"].str.contains(rf"\b{pattern}\b", regex=True)
+                    df[col_name] = pd.Series(pd.NA, index=df.index, dtype="object")
+                    df.loc[matched, col_name] = literal_value
+                else:
+                    df[col_name] = (
+                        df["raw_log"]
+                        .str.extract(rf"\b{pattern}\s+({val_pattern})", expand=False)
+                        .fillna(literal_value)
+                    )
+            else:
+                raise ValueError(
+                    f"Invalid field spec for pattern {pattern!r}: expected 2 or 3 items"
+                )
 
         # Drop verbose columns if not verbose
         if not self._verbose:
@@ -249,6 +271,9 @@ class RcclLogParser:
             "nloops",
             "nsteps",
             "chunksize",
+            "connIndex",
+            "collXhclNum",
+            "p2pXhclNum",
         ]
         for col in numeric_columns:
             if col in df.columns:
@@ -307,6 +332,22 @@ class RcclLogParser:
             sort_cols=["host", "rank", "Pattern"],
         )
 
+    def _report_pcie_mem_channel_info(self):
+        pcie_mem_fields = {
+            r"enable pcie mem channel": ("pcie_mem", None, "enabled"),
+            r"connIndex": ("connIndex", r"\d+"),
+            r"collXhclNum": ("collXhclNum", r"\d+"),
+            r"p2pXhclNum": ("p2pXhclNum", r"\d+"),
+        }
+        self._extract_and_print(
+            title="PCIe Mem Channel Info",
+            filter_func=lambda c: "enable pcie mem channel" in c.lower(),
+            fields=pcie_mem_fields,
+            mandatory=["pcie_mem", "connIndex", "collXhclNum", "p2pXhclNum"],
+            verbose_cols=[],
+            sort_cols=["host", "rank", "connIndex", "collXhclNum", "p2pXhclNum"],
+        )
+
     def _report_collective_transfers(self):
         # Pattern -> column with strict validation
         cl_transfer_fields = {