BatchMatMul2MulFusionPass

Description

MatMul/MatMulv2/MatMulV3/BatchMatmul/BatchMatMulV2/BatchMatMulV3 operators with k=1 exist in the network, and the performance is poor. This graph fusion converts MatMul/MatMulv2/MatMulV3/BatchMatmul/BatchMatMulV2/BatchMatMulV3 into mul to solve the performance problem.

For Ascend 950PR/Ascend 950DT, only the MatMul/MatMulv2/MatMulV3 nodes are fused, not the BatchMatmul/BatchMatMulV2/BatchMatMulV3 nodes.

If the input adj is set to true, insert the reshape operator before the input.

Constraints

• It applies only to static scenarios. The input does not contain bias.
• The dtype meets the following conditions:
  • Float32 for both input and output
  • Float16 or BFloat16 for both input and output

Applicable Products

Atlas A2 training products/Atlas A2 inference products

Atlas A3 training products/Atlas A3 inference products

Ascend 950PR/Ascend 950DT