Skip to content

[Urgent] Repeated consumption of a single logical data in different timesteps #303

@Jackcuii

Description

@Jackcuii

In the latest version (on this branch #293).

In the compiled gemm kernel,

NOT(159) (1,1) (t=11) will send its results to MOV(76) (1,2) (t=12) and MOV(175000) (1,2) (t=13).

MOV(76) and MOV(175000) are both at the northern tile, so 159 will send the result to the NORTH port.

However, the result can not be consumed twice, and it cause the kernel to quit early for exhausting the data.

I am not sure if this issue is related to #293, but maybe we can pend the merge of #293 before we figure out the RC of this issue.

@n0thingNoob Can you paste the IR here for the folks to double check? Thanks~

Metadata

Metadata

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions