[Bug-Report|缺陷反馈]: InplaceAdd、InplaceSub算子迁移融合规则在A3上执行，不合规参数未拦截-ops-nn-AtomGit

已关闭

[Bug-Report|缺陷反馈]: InplaceAdd、InplaceSub算子迁移融合规则在A3上执行，不合规参数未拦截 #4800

梅国晗954517创建于 8月14日关闭于 9月4日

梅国晗954517

8月14日创建

Thanks for sending an issue! Please fill in the following template to help quickly solve your problem.

Describe the current behavior / 问题描述 (Mandatory / 必填)

InplaceAdd、InplaceSub算子在A3上执行，int8/uint8/bf16输入场景下，迁移后融合规则误判生效，导致基础用例执行失败

Environment / 环境信息 (Mandatory / 必填)

1981环境使用主线包

Steps to reproduce the issue / 重现步骤 (Mandatory / 必填)

蓝区A3机器安装主线包，执行相关测试脚本

Describe the expected behavior / 预期结果 (Mandatory / 必填)

不合规参数拦截

[INFO] FE(3672167,python3):2026-07-09-17:00:53.528.296 [fe_ops_kernel_info_store.cc:437]3673381 CheckSupportedByOpsStore:This op[ScatterAdd_0_InplaceAdd_1, ScatterAdd] is not supported. Reason:[tbe-custom]:op type ScatterAdd is not found in this op store.[tbe-custom]:op type ScatterAdd is not found in this op store.[Dynamic shape check]: data type DT_INT8 of input [var] is not supported. All supported data type and format of tensor input0.var is: Data Type: {DT_FLOAT16,DT_FLOAT,DT_INT32,DT_FLOAT16,DT_FLOAT,DT_INT32}Format:{ND,ND,ND,ND,ND,ND}[Static shape check]:data type DT_INT8 of input [var] is not supported. All supported data type and format of tensor input0.var is: Data Type: {DT_FLOAT16,DT_FLOAT,DT_INT32,DT_FLOAT16,DT_FLOAT,DT_INT32}Format:{ND,ND,ND,ND,ND,ND}