siavashzk / branchnet Goto Github PK

Hi, I noticed your instrumentation function in your tracer only records the PC which isn't non-branch, then how did you calculate the following MPKI? I think non-branch instructions should also be considered in this metric.

Why do I prompt 'Unable to load BranchNet/build/tracer/tracer.so: dlopen failed: cannot locate symbol "_ZSt4cerr" referenced by "BranchNet/build/tracer/tracer.so"' when using './bin/create_branch_traces. py'

Hi, may I ask a simple question about your paper? In your paper you said TAGE-SC-L costs 4 cycles to predict, could you please explain why? I can just count 3, maybe I lost some precedures. Thanks!

Hey! I'm trying to reproduce your results but facing the following issue when I try to run the create_branch_traces.py file:

./bin/create_branch_traces.py
Traceback (most recent call last):
File "./bin/create_branch_traces.py", line 52, in
main()
File "./bin/create_branch_traces.py", line 46, in main
cmds.append(get_run_cmd(pinball_path, trace_path))
File "./bin/create_branch_traces.py", line 12, in get_run_cmd
pin_root = os.environ['PIN_ROOT']
File "/usr/lib/python3.6/os.py", line 669, in getitem
raise KeyError(key) from None
KeyError: 'PIN_ROOT'

I have followed the previously required steps. Could you help how to resolve this issue?

@**-Inspiron-3670:/path/to/BranchNet$ ./bin/build_tracer.py
makefile:14: /path/to/BranchNet/src/tracer/source/tools/Config/makefile.config: No such file or directory
make: *** No rule to make target '/media/zhoueg/BranchNet/src/tracer/source/tools/Config/makefile.config'. Stop.

Hello, I have a question here. When I run ./bin/build_tracer.py, an error occurs as the topic shows. I believe I installed Pin Tools, so may I know how to solve this? Looking forward to your reply.

Hello, I am a college student, and our final assignment for computer architecture is to replicate a research paper. I would like to replicate your paper, but I'm facing numerous file configuration issues while using the Intel Pin Tool. Could the author provide the configured files with the generated results and send them to me? The deadline is in only 10 days, and I hope the author can fulfill my request. Thank you very much (T_T(crying)) @siavashzk

Hi, may I know what the weight table stands for in this pic? Is it the weights of the FC layer? Also, I don't understand why there are 41 branch slices, I thought the input was global history so it was the same for every branch.

@siavashzk I was wondering if you could share some examples of code that you used in your paper and explain how to run these through the branchnet training and output. I was looking at the paper and not sure how extreme the examples you used in your code you ran through to get the outputs in your paper were. I am a little lost on how to go from a program to traces to model.

Why does the prompt 'BranchNet/src/tracer/source/tools/Config/makefile.config: No such file or directory' appear when I enter './bin/build_tracer.py'

From the perspective of testing the TAGE + CNN's misprediction rates, are there any specific hardware dependencies -- like a custom processor, required to run this codebase?
I have a Linux system and have installed the following dependencies mentioned in the readme:

Linux (I mainly use CentOS)

Python packages, with the versions that I've used:

Package Version

---

h5py 2.10.0
matplotlib 3.1.1
numpy 1.17.2
PyYAML 5.1.2
torch 1.3.0
torchvision 0.4.1

Intel Pin Tool for generating branch traces (I've tested with 3.5 and 3.11)

cmake3 and make

I have some questions about the dataset construction in our code repository, and I would greatly appreciate your insights. Here are my inquiries:

1. Current Data Set Construction:
As of now, my dataset construction process is as follows:
I have used the leela_s program from spec2017 as an example. Initially, I executed the program with the three default inputs provided by spec2017, namely train, ref, and test, resulting in three traces.
Subsequently, I proceeded to partition the dataset.

However, as per the original research paper's recommendation, I attempted to select "hard-to-predict branches" from the validation set and encountered situations where these branches were not retrievable from the training set. Could you please offer guidance on how to address this issue?
2. Regarding Alberta Inputs:
I've reviewed the scenario where Alberta inputs are used for training. Alberta provides a wide range of inputs. Do I need to generate traces for all of these inputs to use them as part of the training set?
3.About the weight Parameter in benchmark.yaml:
I've noticed that the benchmark.yaml file includes a weight parameter within the simpoint section. I am curious about the significance of this parameter and how one should obtain it.
Thank you in advance for your assistance. Your expertise will be invaluable in resolving these dataset construction queries.