[Mesa-dev] [PATCH] i965: Trim

[Chris Wilson]

Quoting Matt Turner (2018-02-07 20:42:15)
> On Wed, Feb 7, 2018 at 11:41 AM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> >    text    data     bss     dec     hex filename
> > 8852169  388752  671064 9911985  973eb1 lib/i965_dri.so
> > 8852017  388752  671064 9911833  973e19
> > ---
> >  src/mesa/drivers/dri/i965/intel_tiled_memcpy.c | 48 +++++++++-----------------
> >  1 file changed, 16 insertions(+), 32 deletions(-)
> >
> > diff --git a/src/mesa/drivers/dri/i965/intel_tiled_memcpy.c b/src/mesa/drivers/dri/i965/intel_tiled_memcpy.c
> > index 53a5679691..a8157b5232 100644
> > --- a/src/mesa/drivers/dri/i965/intel_tiled_memcpy.c
> > +++ b/src/mesa/drivers/dri/i965/intel_tiled_memcpy.c
> > @@ -468,26 +468,22 @@ linear_to_xtiled_faster(uint32_t x0, uint32_t x1, uint32_t x2, uint32_t x3,
> >        if (mem_copy == memcpy)
> >           return linear_to_xtiled(0, 0, xtile_width, xtile_width, 0, xtile_height,
> >                                   dst, src, src_pitch, swizzle_bit, memcpy, memcpy);
> > -      else if (mem_copy == rgba8_copy)
> > +      else /* if (mem_copy == rgba8_copy) */
> >           return linear_to_xtiled(0, 0, xtile_width, xtile_width, 0, xtile_height,
> >                                   dst, src, src_pitch, swizzle_bit,
> >                                   rgba8_copy, rgba8_copy_aligned_dst);
> > -      else
> > -         unreachable("not reached");
> >     } else {
> >        if (mem_copy == memcpy)
> >           return linear_to_xtiled(x0, x1, x2, x3, y0, y1,
> >                                   dst, src, src_pitch, swizzle_bit,
> >                                   memcpy, memcpy);
> > -      else if (mem_copy == rgba8_copy)
> > +      else /* if (mem_copy == rgba8_copy) */
> >           return linear_to_xtiled(x0, x1, x2, x3, y0, y1,
> >                                   dst, src, src_pitch, swizzle_bit,
> >                                   rgba8_copy, rgba8_copy_aligned_dst);
> > -      else
> > -         unreachable("not reached");
> 
> This should be a no-op. I'd expect the presence of 'else unreachable'
> to allow the compiler to remove the previous conditional.
> 
> >     }
> > -   linear_to_xtiled(x0, x1, x2, x3, y0, y1,
> > -                    dst, src, src_pitch, swizzle_bit, mem_copy, mem_copy);
> > +
> > +   unreachable("not reached");
> 
> I would expect this is what causes the change. If that's the case,
> let's just make this change. What do you think?

Don't mind, I didn't really mean to send this patch (pointed git at the
wrong branch). I was a bit surprised that the compiler treated it any
differently as all branches were terminated by unreachable, and the text
reduction is not enough to account for a linear_to_xtiled() expansion
(they're all meant to be unique expansions aiui).

From a reading pov, having a linear_to_*tiled() after a set of
terminated branches does look a bit odd, so replacing it with an
unreachable() is a good idea.
-Chris