(re-sending from the right email address)
On 2/18/19 12:57 AM, Bryan Erickson wrote:
[...]
> I have modified the make.conf of the build system to include the
following:
>
> CPUTYPE?=sandybridge
> CFLAGS= -mno-aes
>
> However, neither of the changes seem to impact the resulting error
message.
>
> Looking at the build logs, I can see the issue:
>
> This is the Qt Open Source Edition.
> You have already accepted the terms of the Open Source license.
> Running configuration tests...
> Checking for valid makespec... yes
> Checking for target architecture... x86_64
> Checking for SSE2 instructions... yes
> Checking for AES new instructions... yes
[...]
> But it does see my custom CFLAGS and CPUType:
[...]
> Would you have any advice on how to proceed?
>
> Is there an additional modification that I need to my make.conf or is
there some method for the build process to override the checks if the
CFLAGS state no AES or the CPUTYPE is of a processor which doesn't have
AES support?
If I understood you correctly, you're building the packages on a machine
that does support AES, but you have control over CPUTYPE and other flags
on a per-target basis, so that you can set them to a different value for
your SandyBridge machine that doesn't support AES, right?
You essentially need to pass -no-aesni to Qt's configure script. We
already do something similar in spirit to disable SSE2 on i386 when it's
not supported (see Mk/Uses/qt-dist.mk). I think the quickest option for
you would be patching that file and adding a check for
CPUTYPE=="sandybridge" and passing that extra argument in CONFIGURE_ARGS.
We could do that in the ports tree, but there doesn't seem to be an
entry for aesni we could check in bsd.cpu.mk as far as I can see.
