ASSERT FAILED rid >= 0

[wonjunsong]

Hi all.

when i simulated 4 cpu spec2006 application(leslie3d,lbm,cactusADM,sjeng) and 1 gpu rodinia applicatiion(nw), there was a assertion below.

src/dram.cc:396: ASSERT FAILED (I=50000000  C=398473490):  rid >= 0
src/dram.cc:396: ASSERT FAILED (I=50000000  C=398473490):  addr:800bffff cid:1792 bid:5 rid:-2146697217 type:IFETCH

i use "params_hetero_4c_6g".

what is the problem? the gpu trace problem??

any advice please.

Regards.

[이성광]

I had that problem. The local variable rid is declared as a int, but it can be more than 2^31.

I solved that problem by changing the type of rid to int64.

2013/1/23 wonjunsong <iamson...@gmail.com>

--
 
 
 

[wonjunsong]

Thanks. 

[wonjunsong]

should i change the type of all related variable ?? such as rid, cid, bid, bid_xor, m_bid, m_rid, m_cid ?? right?

[Jaewoong Sim]

In fact, if everything is correct, "int rid" should be OK since we use a 32-bit address. Although a memory request is like 0x80000000, rid is shifted right by some values, so that rid never can have a minus value. So, the fact that you had the error means something is wrong.

Fortunately, I was able to reproduce the exact same error with no the following configuration. Note that this is a wrong configuration since there is no GPU core. But, the code should be more robust, so we will fix this.. If the repeat trace is 0, the simulation is also fine.

Thanks,
Jaewoong

--------------------------------------------------------------------------------------

1. trace_file_list has 4 cpu + 1gpu apps

2. NO GPU cores & repeat traces

num_sim_cores 4

num_sim_small_cores 0

num_sim_large_cores 4

max_insts 200

repeat_trace 1

---

src/process_manager.cc:495: (I=200  C=5102072):  thread_count:1

**Core 0 Thread 305 Finished:   insts:200         cycles:638770      seconds:9     -- 0.00 IPC (0.00 IPC) --  N/A  KHz (0.02 KHz) 

src/process_manager.cc:433: (I=200  C=5110152):  setup_process:1 /trace/x86/spec2006/trace_simpoint/mcf/pin_traces/mcf.1.txt current_index:0 (1)

src/process_manager.cc:495: (I=200  C=5110152):  thread_count:1

src/dram.cc:343: ASSERT FAILED (I=200  C=5112030):  rid >= 0

src/dram.cc:343: ASSERT FAILED (I=200  C=5112030):  addr:802000ff cid:192 bid:6 rid:-2145386241 type:IFETCH

--
 
 
 

[Jaewoong Sim]

I was able to reproduce this error with having GPU cores as well. So, with params_hetero_4c_4g, when specifying 4 CPU applications and 6 GPU applications in trace_file_list and repeats the traces (repeat_trace=1), I got the same error. If I do not repeat the traces, only one GPU application is running and the simulation is done. (This is because all the GPU cores are allocated to the one GPU application, and the simulation is finished without running the next GPU applications in the trace_file. I did not implement this part, so I have no idea whether this was intended behavior.)

If you want to run each GPU application in the trace_file_list on each different GPU core (this is probably YOUR intention), please set "max_num_core_per_appl" to 1. That will fix the issue.

Thanks,

Jaewoong

[wonjunsong]

I just want to simulate 4 cpu application and only 1 gpu application(also want that 1 gnu application is assigned to all gpu core)

I had a question about this before .

(https://groups.google.com/forum/?fromgroups=#!searchin/macsim-dev/max_num_core_per_appl/macsim-dev/0wY-914MkTc/HZme3FVI9SUJ)

[王庆林]

To wonjunsong :
    I encounter the same question with you now.Could you tell me how to resolve it yourself finally?
Thank you!

2013/1/26 wonjunsong <iamson...@gmail.com>

--
 
 
 

[Jaekyu Lee]

Hi,

We've fixed the problem, but this change is not included in the any version of tags.

However, the trace format has been changed recently.

As a result, you have to generate traces again to use the latest version of macsim.

Thanks,

Jaekyu

--
 
---
You received this message because you are subscribed to the Google Groups "Macsim Developer" group.
To unsubscribe from this group and stop receiving emails from it, send an email to macsim-dev+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[王庆林]

To Jaekyu:
   Thank you. 

   I got the macsim version 392 and GPUtrace x86TraceGenerator.cpp r2117.When doing simulation,there was a assertion below.

src/process_manager.cc:385: ASSERT FAILED (I=0  C=0):  0
src/process_manager.cc:385: ASSERT FAILED (I=0  C=0):  this version of the simulator supports only version 1.31 of the GPU traces
Command exited with non-zero status 15

   Could you tell me how to get the GPU traces of version1.31？
  
   Thanks.
   Bigpig

2013/5/8 Jaekyu Lee <jq.l...@gmail.com>

[Jaekyu Lee]

Hi,

Could you try this again in a few days?

We made major changes and testing is in progress now.

This will be done in a few days and everything should be up at that time.

We are sorry for any inconvenience.

Thanks,

Jaekyu

[wonjunsong]

Hi Jaekyu.

how about the progress??

Thanks for your effort :-))

[Jaekyu Lee]

Hi Wonjun,

You may try with the latest version.

I have fixed some bugs.

This change may fix the problem with the prefetcher.

However, you may need to regenerate traces.

Thanks,

Jaekyu

[wonjunsong]

Thanks a lot.

I'll try soon.

Thanks :-)