Comment author: @nojb

Added file daxpy.s.diff showing the diff between the generated assembly with -malign-double and -mno-align-double. You can see there that "aligned" instructions movaps and movapd are replaced by "unaligned" instructions movups and movupd.

Comment author: @nojb

Also, you can find out your default -malign-double gcc setting by inspecting

gcc -Q --help=target

Comment author: @xavierleroy

There is a configure-time test to try and see whether doubles need 8-alignment, see config/auto-aux/dblalign.c. However it is easily fooled by clever compilers, hence the many exceptions in the configure script.

This issue has been open one year with no activity. Consequently, it is being marked with the "stale" label. What this means is that the issue will be automatically closed in 30 days unless more comments are added or the "stale" label is removed. Comments that provide new information on the issue are especially welcome: is it still reproducible? did it appear in other contexts? how critical is it? etc.

This looks like a bug in the attached C stub, which directly casts an OCaml value to double*:

      daxpy(Wosize_val(a), (double *)a);

Accessing floats from C should be instead done using the Double_val macro that the FFI provides, which expands to different code depending on whether ARCH_ALIGN_DOUBLE is set. (This setting is detected more reliably since #8532)

Thanks @stedolan for the explanation, sounds right!