Vector Bottleneck Operators

{% if rank is not none %} Analysis of rank {{ rank|safe }}. {% endif %}
Description Suggestion
{{ format_result.description|safe }} {{ format_result.suggestion|safe }}
{% for op_type, op_info in format_result.statistic %}
{{ op_type|safe }}
Operator Type Counts Elapsed Time(us) Compution Ratio(%) Step Time Ratio(%)
{{ op_info.summary.op_type|safe }} {{ op_info.summary.counts|safe }} {{ op_info.summary.total_duration|safe }} {{ op_info.summary.op_computation_ratio|safe }} {{ op_info.summary.op_step_time_ratio|safe }}

Suggestion: {{op_info.summary.suggestion|safe}} {% if op_info.summary.op_priority == 2 %} {% if language == 'en' %} current CPU and communication time relatively high, Please check the related problem. {% else %} 当前CPU和通信时间占用较高,请检查相关问题。 {% endif %} {% endif %}

{% for _, stacks in op_info.op_info_list.items() %} {% if not stacks.op_info_list[0].stack_info %} Suggestion: {% if language == 'en' %} These APIs have no stack. If parameter 'with_stack=False' was set while profiling, please refer to Ascend Profiler to set 'with_stack=True'. {% else %} 当前API没有堆栈信息。如果在性能分析时设置了参数'with_stack=False',请参考 Ascend Profiler 来设置'with_stack=True'。 {% endif %} {% else %}
No.{{loop.index|safe}} code stack, input shapes is: {{stacks.op_info_list[0].input_shapes}}, called {{stacks.summary.counts|safe}} times
{% endif %} {% endfor %}
{% endfor %}