gpu-compute: Fixed a bug in global memory pipeline
Review Request #3470 - Created May 18, 2016 and updated
| Information | |
|---|---|
| Tuan Ta | |
| gem5 | |
| Reviewers | |
| Default | |
gpu-compute: Global memory pipeline should not increment inflightStores counter for memory fence instruction. Deadlock could happen when there are many fence instructions executed by a compute unit because inflightStores will eventually go over gmQueueSize.
Ship It!
Ship It!
Could you make sure the summary is <65 char, and add a somewhat more elaborate description? Thanks.
Summary: |
|
||||||
|---|---|---|---|---|---|---|---|
Description: |
|
The description still needs to be fixed. It should follow this form:
gpu-compute: summary
description
Also, this is not a VIPER specific bug, it's a bug in the way the global pipeline handles these counts.
Summary: |
|
||||||
|---|---|---|---|---|---|---|---|
Description: |
|
Thanks, Tuan. I will ship this soon.