[Mesa-dev] merged/mega *_dri.so files?

[Emil Velikov]

On 9 March 2017 at 09:35, Steven Newbury <steve at snewbury.org.uk> wrote:
> On Wed, 2017-03-08 at 21:29 -0700, Brian Paul wrote:
>> I just noticed something:
>>
>> $ ls -l lib/*_dri.so
>> -rwxr-xr-x 8 brianp users 43779971 Mar  8 20:59 lib/i915_dri.so*
>> -rwxr-xr-x 8 brianp users 43779971 Mar  8 20:59 lib/i965_dri.so*
>> -rwxr-xr-x 8 brianp users 43779971 Mar  8 20:59
>> lib/nouveau_vieux_dri.so*
>> -rwxr-xr-x 8 brianp users 43779971 Mar  8 20:59 lib/r200_dri.so*
>> -rwxr-xr-x 8 brianp users 43779971 Mar  8 20:59 lib/radeon_dri.so*
>> -rwxr-xr-x 8 brianp users 43779971 Mar  8 20:59 lib/swrast_dri.so*
>>
>> $ ls -l lib/gallium/*_dri.so
>> -rwxr-xr-x 8 brianp users 47831465 Mar  8 20:59
>> lib/gallium/i915_dri.so*
>> -rwxr-xr-x 8 brianp users 47831465 Mar  8 20:59
>> lib/gallium/kms_swrast_dri.so*
>> -rwxr-xr-x 8 brianp users 47831465 Mar  8 20:59
>> lib/gallium/nouveau_dri.so*
>> -rwxr-xr-x 8 brianp users 47831465 Mar  8 20:59
>> lib/gallium/r300_dri.so*
>> -rwxr-xr-x 8 brianp users 47831465 Mar  8 20:59
>> lib/gallium/swrast_dri.so*
>> -rwxr-xr-x 8 brianp users 47831465 Mar  8 20:59
>> lib/gallium/virtio_gpu_dri.so*
>> -rwxr-xr-x 8 brianp users 47831465 Mar  8 20:59
>> lib/gallium/vmwgfx_dri.so*
>>
>> All the drivers are the same size.  In fact,
>>
>> $ md5sum lib/gallium/*_dri.so
>> ee2e5c4dc84001e99f54a6ddb8a5ee88  lib/gallium/i915_dri.so
>> ee2e5c4dc84001e99f54a6ddb8a5ee88  lib/gallium/kms_swrast_dri.so
>> ee2e5c4dc84001e99f54a6ddb8a5ee88  lib/gallium/nouveau_dri.so
>> ee2e5c4dc84001e99f54a6ddb8a5ee88  lib/gallium/r300_dri.so
>> ee2e5c4dc84001e99f54a6ddb8a5ee88  lib/gallium/swrast_dri.so
>> ee2e5c4dc84001e99f54a6ddb8a5ee88  lib/gallium/virtio_gpu_dri.so
>> ee2e5c4dc84001e99f54a6ddb8a5ee88  lib/gallium/vmwgfx_dri.so
>>
>> they're identical.  I don't recall any of the proposed "mega driver"
>> stuff landing.  Can someone explains what's going on?
>>
>
> As was mentioned, it landed long ago, but why is your driver so huge?
> Unstripped? Is LTO making *that* much difference?  I'm not optimizing
> for size.
>
The ones listed by Brian are not stripped.

In my earlier testing 50-90% of the binary size was the shared/common code.
That said - if we have a compelling reason for a
"--build-separate-drivers" ~like configure toggle, I won't object.

-Emil