Quoting Sanjay Singh <[EMAIL PROTECTED]>: > With your changes added in, it appears to behave like version .25 did.
Good news! > Before, with GHDL .27, it would crash with the Bus Error. Will this change > work with Linux as well? I hope it does. Yes. > Now, Tristan, please take a couple of minutes and help me/us understand > better how you figured out that this bit of code would likely fix things. Yes, I wrote my previous mail in the hurry. > How long did it take you? What sorts of things did you consider? In fact this comes from one of your previous mail where you gave both crash location and corresponding assembly code. >From the assembly code, it was clear that the processor did an unaligned access, which is illegal on SPARC but not on x86. > I was surmising that it was the Trans.Val_Ptr that was the problem, but I > would not have known what to change or how to make an elegant solution > work. > > The ldd assembly language instruction and the values pulled from it via > gdb implied that something was not 32-bit aligned, because one of those > values when divided mod 4 gave a result with a remainder of .5 Yes, this was the hint. But maybe you didn't know that I did a block copy which is not fully correct. >... so > something seemed amiss there, but its clear you are at a level of skill > thats extraordinary. Well, with more skill I had avoided this bug :-) > In order for any of us to lend a hand at the lower levels of GHDL in > future, we need to have some idea of what kinds of skills are needed to > get up to where we can be of some assistance to you and the GHDL project. > At least I am of this sentiment. My TODO list: * Add more vhdl 2008 features. This requires a good knowledge of VHDL, and will require Ada skills to implement these features. Some of them should be rather easy to implement (only some local changes), some of them can be very complex. * Improve simulation speed. First we need to understand what is slow... If you have good profiling skills, you're welcome. * Use LLVM as a back-end. This doesn't require a good Ada knowledge but LLVM and how a simple compiler (GHDL/Ortho in this case) works. > If you can spare a little time, please inform us of the process by which > you arrived at the fix. You did this even without a Solaris environment > which makes this all the more remarkable. As I said, just seeing the assembly code (and you did a very good job at sending the right part) was enough, given that I know this part well. > I will rebuild everything from scratch with the 4.1.2 compiler, optimize > the binaries, strip them, and do a few small Solaris tweaks needed (cause > Solaris 8 lacks some Linux functions) and then send the whole thing up to > you for installation on the download page of your site. Good. Being able to build GHDL is still not very easy. > Honestly, I'm blown away by your talents. I'll get back to you with any > other information after I've had a chance to test the Solaris port. Ok, tell us the results. > Lastly, before I forget, there was a page that I read about using > Solaris's built-in threading libraries to help with making use of more > processors. Thats something that if its possible to test, would be a big > win for the Solaris port. I'll find that page and would like to get your > opinion of it. Ok. I have to revive the initial multi-threaded kernel work. Might be a good start if you want to work on it. Thank you for your patience and your work, Tristan. _______________________________________________ Ghdl-discuss mailing list [email protected] https://mail.gna.org/listinfo/ghdl-discuss
