Won't help, he needs to update binutils. Even if he uses -march=core2 it will still emit AVX2 instructions and old GNU assembler will break the compilation -- the file responsible is algo/argon2/ar2/sj/scrypt-jane-mix_salsa64-avx2.h, it will emit that instruction no matter what -march compiler option is provided.

You can easily replicate that problem yourself with installing gcc 4.8 and binutils 2.20 (which doesn't support avx2) in a chroot, container or virtual machine.

GNU assembler is too old as well, unfortunately. It doesn't know what 'vpermq' is -- that's an AVX2 instruction.

Same goes for other instructions it doesn't know about.

GNU assembler is part of binutils. Try installing newer version of binutils.

I know for sure that binutils 2.25 supports it, but some earlier versions likely will too.

This is different error, and you didn't include it here, it's higher. You've pasted make output saying that there was an error above the pasted text.

You need 4.5 or newer.

Yes, you need to upgrade to newer gcc. Or ask joblo to support a compiler that's 9 years old.

You're likely using gcc 4.4. Use newer compiler.

This particular one you quoted -- your compiler options did not enable SSSE3 support. Please use -march=native instead of -mtune=broadwell

That's only source code there.

Thanks!

If anyone in this thread is interested, I'm hosting those at https://hmage.net/cpuminer-opt/ — the link is permanent.

Wasn't paying attention to the project for some time, so missed a big amount of the releases. And old links posted here are no longer working.

I'd like the older ones, and looks like th3.root doesn't have 3.4.5 version in his archive.

Do you maybe have a single archive of all of the versions?

Would be very grateful ^.^

Thanks! Sent the PM.

Hey joblo, do you keep archive of previous versions anywhere? I'm doing research on incremental benchmarks between versions for archival purposes.

Would be grateful if you could give access to previous .tar.gz releases all the way back to the first one.

Thanks in advance.

QEMU supports AES, but by default it shows a minimal subset of CPU features so that the VM can be moved between physical machines and compiled code on one host will run on another host.

Add "-cpu host" to qemu commandline. But beware that the VM will not run properly on another machine that doesn't support same CPU features if you compile anything with "-march" flag.

[EDIT]: Also I don't see KVM enabled. Enable KVM with "-enable-kvm" qemu flag in command line.

Your wallet.

If he's compiling for AES he shouldn't even need to have non-AES version compiled at all.

Good coding style. Will accept that in my repo.

[EDIT]: merged.

For the future, gcc has a flag: -fstack-usage

when compiling it generates *.su files that have info how much bytes a function would need for stack.

After adding that, recompiling cpuminer-opt on core2 and then printing all *.su files, sorted by size:


You might want to increase max stack size in makefile to 3mb or more, setting it to 2MB isn't enough because you need 2097648 bytes just for that function.


"-Wl," means "pass this to linker", which means ld, so you need to check the documentation of ld -- http://linux.die.net/man/1/ld

On Windows, stack size limit is specified in the binary. On Linux, the limit is set by system administrator. On current debian, default is 8Mb.


This is different feature from above, and -fno-stack-limit is used to negate -fstack-limit-register/-fstack-limit-symbol, by default this feature is not set.

HODL

msys comes with gdb, it should be able to catch the segfault, then you can inspect the registers, stack and variables.

Compiling with "-O0 -g3" instead of "-O3" should help gdb give you more info.

It means that your build of gcc doesn't have stack protection support library (functions __stack_chk_fail() and __stack_chk_guard()). It builds fine on debian linux though.