sarchlab / mgpusim

A highly-flexible GPU simulator for AMD GPUs.
MIT License
62 stars 9 forks source link

VALU instructions should take 4 cycles rather than 5 cycles. #29

Open syifan opened 7 months ago

syifan commented 7 months ago

To Reproduce MGPUSim version of commit ID: 7810efc

Command that recreates the problem

./fir -timing --trace-vis

Current behavior It takes 5 cycles for VALU instructions to execute. And because of the round-robin scheduler, the gap between two instructions are 8 cycles.

Expected behavior If there is only one wavefront running in one CU and there is only VALU instructions, VALU instructions should lay back to back.