Error building X86_SE: ‘class X86ISA::DTB’ has no member named ‘flushAddr’

[Ankita (Garg) Goel]

Hi,

I am trying to build the simulator for X86_SE. However, I get the following error:

build/X86_SE/cpu/simple_thread.cc: In member function ‘void SimpleThread::deallocateStack()’:

build/X86_SE/cpu/simple_thread.cc:339: error: ‘class X86ISA::DTB’ has no member named ‘flushAddr’

cc1plus: warnings being treated as errors

build/X86_SE/sim/host.hh: At global scope:

build/X86_SE/sim/host.hh:61: warning: ‘MaxTick’ defined but not used

scons: *** [build/X86_SE/cpu/simple_thread.fo] Error 1

scons: building terminated because of errors.

I did look at the sources and found that in X86, flushAddr has not been defined as a member of the TLB class (unlike in the case of alpha). Has anyone come across this error ? Anyone has any suggestions on how we could add the flush routine to the X86 TLB ?

Thanks for your help !

--
Regards,
Ankita

[Jiayuan Meng]

Right now MV5 only supports Alpha, because the threading library in SE mode is based on Alpha ISA. 

Jiayuan

[Ankita (Garg) Goel]

Ok. I am looking for heterogeneous architecture support on x86. I want to evaluate the benefits of working on extending MV5 to x86. So in this regard, wanted to understand the following:

- In your experiments, what is the largest SIMD & OOO system configuration you could run without affecting the performance of simulation ? i.e, simulation not being too slow ?

- What is the typical time it takes to run one of the fractal/RODINIA benchmarks on a heterogeneous simulation ?

Answers to the above questions will greatly help me. Appreciate your response.

Regards,

Ankita

--
Regards,
Ankita

[Jiayuan Meng]

- In your experiments, what is the largest SIMD & OOO system configuration you could run without affecting the performance of simulation ? i.e, simulation not being too slow ?

I've run up to 256 SIMD cores, each with a SIMD width of 4 and 2 warps, together with 2 OOO cores, 

- What is the typical time it takes to run one of the fractal/RODINIA benchmarks on a heterogeneous simulation ?

It ranges from 10 minutes to 2 days, but for the kernels which I usually play around with, they finish mostly within half a day or a few hours. It also depends on the input size.