Remove unnecessary memcpy when reading WAL record fitting on page

When reading a WAL record, its contents are copied into an intermediate
buffer.  However, doing so is not necessary if the record fits fully
into the current page, saving one memcpy for each such record.  The
allocation handling of the intermediate buffer is also now done only
when a record crosses a page boundary, shaving some extra cycles when
reading a WAL record.

Author: Andrey Lepikhov
Reviewed-by: Kyotaro Horiguchi, Heikki Linnakangas
Discussion: https://postgr.es/m/c2ea54dd-a1d3-80eb-ddbf-7e6f258e615e@postgrespro.ru

src/backend/access/transam/xlog.c

@@ -4203,9 +4203,6 @@ CleanupBackupHistory(void)
  * If no valid record is available, returns NULL, or fails if emode is PANIC.
  * (emode must be either PANIC, LOG). In standby mode, retries until a valid
  * record is available.
- *
- * The record is copied into readRecordBuf, so that on successful return,
- * the returned record pointer always points there.
  */
 static XLogRecord *
 ReadRecord(XLogReaderState *xlogreader, XLogRecPtr RecPtr, int emode,

src/backend/access/transam/xlogreader.c

@@ -353,19 +353,6 @@ XLogReadRecord(XLogReaderState *state, XLogRecPtr RecPtr, char **errormsg)
 		gotheader = false;
 	}
 
-	/*
-	 * Enlarge readRecordBuf as needed.
-	 */
-	if (total_len > state->readRecordBufSize &&
-		!allocate_recordbuf(state, total_len))
-	{
-		/* We treat this as a "bogus data" condition */
-		report_invalid_record(state, "record length %u at %X/%X too long",
-							  total_len,
-							  (uint32) (RecPtr >> 32), (uint32) RecPtr);
-		goto err;
-	}
-
 	len = XLOG_BLCKSZ - RecPtr % XLOG_BLCKSZ;
 	if (total_len > len)
 	{
@@ -375,6 +362,19 @@ XLogReadRecord(XLogReaderState *state, XLogRecPtr RecPtr, char **errormsg)
 		char	   *buffer;
 		uint32		gotlen;
 
+		/*
+		 * Enlarge readRecordBuf as needed.
+		 */
+		if (total_len > state->readRecordBufSize &&
+			!allocate_recordbuf(state, total_len))
+		{
+			/* We treat this as a "bogus data" condition */
+			report_invalid_record(state, "record length %u at %X/%X too long",
+								  total_len,
+								  (uint32) (RecPtr >> 32), (uint32) RecPtr);
+			goto err;
+		}
+
 		/* Copy the first fragment of the record from the first page. */
 		memcpy(state->readRecordBuf,
 			   state->readBuf + RecPtr % XLOG_BLCKSZ, len);
@@ -479,7 +479,6 @@ XLogReadRecord(XLogReaderState *state, XLogRecPtr RecPtr, char **errormsg)
 		state->EndRecPtr = RecPtr + MAXALIGN(total_len);
 
 		state->ReadRecPtr = RecPtr;
-		memcpy(state->readRecordBuf, record, total_len);
 	}
 
 	/*