scripts: update converter with quantization mode

scripts/verbose_converter/src/benchdnn_generator.py

@@ -195,6 +195,8 @@ class Converter(metaclass=ConverterMeta):
             # Set policy to "host_scalar" if is_host_scalar is True
             if param.is_host_scalar:
                 policy = "host_scalar"
+            if param.quantization_mode == "dynamic_mx":
+                policy = "mx"
             result = f"{arg}:{policy}"
             if policy == "common" or policy == "host_scalar":
                 result += f":{def_value}"

scripts/verbose_converter/src/ir.py

@@ -307,6 +307,7 @@ class QuantizationParam(Mapping):
     mask: int = 0
     groups: str = ""
     is_host_scalar: bool = False
+    quantization_mode: str = ""
 
     def __str__(self):
         if self.groups:

scripts/verbose_converter/src/parse.py

@@ -411,7 +411,7 @@ class ParserImpl:
     @staticmethod
     def parse_quantization_param(spec, read_value, param_type):
         # Old style: mask[:[value[*]|*]]
-        # New style: mask[:data_type[:groups]]
+        # New style: mask[:data_type[:host_scalar[:groups[:quantization_mode]]]]
         param = param_type()
         param.mask = spec.read_uint()
         if spec.read_literal(":"):
@@ -429,6 +429,9 @@ class ParserImpl:
                         param.is_host_scalar = True
                     else:
                         param.groups = groups_or_host_flag
+                    if spec.read_literal(":"):
+                        param.quantization_mode = spec.read_str()
+
         return param
 
     # v2.7 and below