Are you sure?  From the log output you posted above, and from your comments, I thought it wasn't working for you?

Both the 'dag-export' version that you have already used and the other 'generate-dag' branch that I mentioned are based off of ethminer-0.9.41-genoil-1.1.7, so maybe that wouldn't be the best for you.  What version of ethminer generally works the best?  Genoil's?  NerdRalph's?   I solo mine so I am not well informed about the stratum issues with Genoil's miner.  I will see if I can get a build uploaded for that 'generate-dag' branch soon.

I compared the dag file you uploaded and it matches perfectly with the official version, so that means you are not affected by this issue.  So you don't need the fixed version.  You can continue using what you were using before, or use ethminer-0.9.41-genoil-1.1.7, which is what I based my work off of.  If you really want a fixed version, you can download my generate-dag branch and build from source.

Looks like you've got the -G parameter twice.

Ok, do it that way if you want.  Just so were clear, you're going to be running the modified Genoil ethminer that I posted, and you're going to post a link to dag file that it produces, which is dag_Genoil.dat. Make sure you also let me know what it says when ethminer first starts up ... something like  Initialising miner for xxxx .  I need to know what the "xxxx" is.  And also what OS and driver version you are running.

Well, the file that will be produced will be a full DAG file, which currently runs at about 2GB, so I don't think you will want to upload that anywhere. 

The idea is you run the original Ethereum Foundation ethminer, which will create a DAG file which you will use as the standard of comparison.  Then you run the modified ethminer that I've provided which creates the DAG in the GPU memory as usual, and then dumps it out to a separate file.  Then you use a file compare utility to see if the two files are the same.  You can post the output of the file compare utility here if you want, or just indicate whether the two files are the same or not.

The GitHub page I linked to has detailed instructions.

Hello,

By a bit of a coincidence I have discovered that on my machine at least, the last 2176 bytes of the DAG file are not being generated properly (they are zero). This corresponds to the last 34 dag entries. This is on Windows 10. The DAG entries are used to compute hashes, which is integral to the mining process. If the DAG entries are wrong, it will affect your mining results. To put it in perspective, though, the DAG currently contains about 33 million entries, so this issue will have a pretty small impact.

I don't think there is anything wrong with @genoil's code ... at least not that I can see. It's either a bug in OpenCL, or something specific to my machine / GPU, which is a Gigabyte R9 380X 4GB.  I'm running with AMD drivers 16.12.2.  I was unable to reproduce this issue running under Linux.

I thought it would be important to determine if this is a more widespread issue, so I forked Genoil's ethminer and added some code to export the DAG to a file, so it can be compared to the same DAG file created by the original EF Ethminer.  (Note: I had already tried submitting an issue / PR on Genoil's GitHub site, but as he has said here recently, he's not actively developing any more.  From what I can see, though, he does pop up here occasionally.)

If anyone wants to try to reproduce this issue you can visit my GitHub site and read more about it.  If you report your results, please include which AMD driver version you are using.