On 10/13/21 10:13 AM, Christoph Hellwig wrote:

On Wed, Oct 13, 2021 at 10:04:36AM -0600, Jens Axboe wrote:

quoted

On 10/13/21 9:50 AM, Christoph Hellwig wrote:

quoted

On Wed, Oct 13, 2021 at 09:42:23AM -0600, Jens Axboe wrote:

quoted

Something like this?

Something like that.  Although without making the new function inline
this will generate an indirect call.

It will, but I don't see how we can have it both ways...

Last time I played with these optimization gcc did inline function
pointers passed to __always_inline function into the calling
function.  That is you can keep the source level abstraction but get the
code generation as if it was open coded.

Gotcha, so place nvme_complete_batch() in the nvme.h header. That might
work, let me give it a whirl.

-- 
Jens Axboe