The Peformance Problem of Boom

link liao

unread,

Oct 20, 2019, 11:54:00 PM10/20/19

to riscv-boom

Hi, I am new to Boom, last week, I use config of WithMediumBooms to generate Boom RTL, and want to match Cortex-A9,But

when I run the program coreMark at Boom, it just only about 2.24 Coremark/MHz, and during runing the program,I find that the branch misprediction rate is very high, how can I do to fix this bug

Jerry Zhao

unread,

Oct 20, 2019, 11:58:18 PM10/20/19

to link liao, riscv-boom

We are aware of the branch misprediction issues in BOOM right now, and are currently working to address them.
LargeBoom and MegaBoom will perform better currently.

On Sun, Oct 20, 2019 at 8:54 PM link liao <linkin...@gmail.com> wrote:

Hi, I am new to Boom, last week, I use config of WithMediumBooms to generate Boom RTL, and want to match Cortex-A9,But
when I run the program coreMark at Boom, it just only about 2.24 Coremark/MHz, and during runing the program,I find that the branch misprediction rate is very high, how can I do to fix this bug

--
You received this message because you are subscribed to the Google Groups "riscv-boom" group.
To unsubscribe from this group and stop receiving emails from it, send an email to riscv-boom+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/riscv-boom/9815cfed-1657-4d61-8869-25f513b223bc%40googlegroups.com.

Farzad Farshchi

unread,

Nov 3, 2019, 3:06:47 PM11/3/19

to Jerry Zhao, riscv-boom

Hi Jerry,

I get 2.72 Coremark/MHz with LargeBomm running on the latest version of FireSim. Does LareBoom have the same branch misprediction problem? Is there a quick fix to this issue?

Thanks,

Farzad

To view this discussion on the web visit https://groups.google.com/d/msgid/riscv-boom/CAC%2BpDSH%2B26Ph2W8-EOpP3h_vg%3DYqm2cqj1zPupwX7x_WaTpMCw%40mail.gmail.com.

ZiXin Guan

unread,

Feb 23, 2020, 9:25:59 PM2/23/20

to riscv-boom

Hi Jerry,

I get 2.87Coremark/Mhz with the latest version of chipyard (Feb,14,2020), Is there any old version for us to get 4.7 Coremark/MHz?
Thanks,

Gibson

在 2019年10月21日星期一 UTC+8上午11:58:18，Jerry Zhao写道：

We are aware of the branch misprediction issues in BOOM right now, and are currently working to address them.
LargeBoom and MegaBoom will perform better currently.

On Sun, Oct 20, 2019 at 8:54 PM link liao <linkin...@gmail.com> wrote:

Hi, I am new to Boom, last week, I use config of WithMediumBooms to generate Boom RTL, and want to match Cortex-A9,But
when I run the program coreMark at Boom, it just only about 2.24 Coremark/MHz, and during runing the program,I find that the branch misprediction rate is very high, how can I do to fix this bug

--
You received this message because you are subscribed to the Google Groups "riscv-boom" group.

To unsubscribe from this group and stop receiving emails from it, send an email to riscv...@googlegroups.com.

Jerry Zhao

unread,

Feb 23, 2020, 9:33:27 PM2/23/20

to ZiXin Guan, riscv-boom

Yes, but I don't know the exact commit. You'll need to track down old compatible versions of Chisel, FIRRTL, and RocketChip. Also we don't back port bug fixes, so older versions will be buggy. I don't recommend trying to get an older version to work.

We will release a newer version of BOOM soon with performance exceeding any prior version.

To unsubscribe from this group and stop receiving emails from it, send an email to riscv-boom+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/riscv-boom/a8bf6f88-5458-4f7c-87cd-4344dc2b2e9d%40googlegroups.com.

Gibson Guan

unread,

Mar 2, 2020, 4:49:49 AM3/2/20

to riscv-boom

Thanks Jerry.

That sounds great for us.

Do you have the schedule of the new version?

在 2020年2月24日星期一 UTC+8上午10:33:27，Jerry Zhao写道：

To view this discussion on the web visit https://groups.google.com/d/msgid/riscv-boom/a8bf6f88-5458-4f7c-87cd-4344dc2b2e9d%40googlegroups.com.

Reply all

Reply to author

Forward