[Mesa-dev] [PATCH] glsl: Optimize logic operation A || (A && B)

[Thomas Helland]

11. juli 2014 01:02 skrev "Jordan Justen" <jljusten at gmail.com> følgende:
>
> On Thu, Jul 10, 2014 at 2:24 PM,  <thomashelland90 at gmail.com> wrote:
> > From: Thomas Helland <thomashelland90 at gmail.com>
> >
> > Let's cut the needless A && B here.
> > Gives some effect on a clean shader-db with
> > some extra shaders from TF2 and portal.
> >
> > helped: shaders/tf2/2042.shader_test fs16:                23 -> 21
> > (-8.70%)
> > helped: shaders/tf2/2042.shader_test fs8:                 23 -> 21
> > (-8.70%)
> > helped: shaders/tf2/4624.shader_test fs16:                21 -> 19
> > (-9.52%)
> > helped: shaders/tf2/4624.shader_test fs8:                 21 -> 19
> > (-9.52%)
> > helped: shaders/tf2/763.shader_test fs16:                 23 -> 21
> > (-8.70%)
> > helped: shaders/tf2/763.shader_test fs8:                  23 -> 21
> > (-8.70%)
> >
> > HURT:   shaders/orbital_explorer.shader_test vs:          1049 -> 1052
> > (0.29%)
> >
> > total instructions in shared programs: 758979 -> 758970 (-0.00%)
> > instructions in affected programs:     1183 -> 1174 (-0.76%)
> > GAINED:                                0
> > LOST:                                  0
> >
> > Signed-off-by: Thomas Helland <thomashelland90 at gmail.com>
> > ---
> >  src/glsl/opt_algebraic.cpp | 10 ++++++++++
> >  1 file changed, 10 insertions(+)
> >
> > diff --git a/src/glsl/opt_algebraic.cpp b/src/glsl/opt_algebraic.cpp
> > index ac7514a..8f3a505 100644
> > --- a/src/glsl/opt_algebraic.cpp
> > +++ b/src/glsl/opt_algebraic.cpp
> > @@ -588,6 +588,16 @@
ir_algebraic_visitor::handle_expression(ir_expression *ir)
> >        } else if (ir->operands[0]->equals(ir->operands[1])) {
> >           /* (a || a) == a */
> >           return ir->operands[0];
> > +      } else if (op_expr[0] && op_expr[0]->operation ==
ir_binop_logic_and &&
> > +                    (op_expr[0]->operands[0]->equals(op_expr[1]) ||
> > +                     op_expr[0]->operands[1]->equals(op_expr[1]))) {
> > +         /* A || (A && B)  or A || (B && A) */
> > +         return ir->operands[0];
>
> Isn't this returning the 'and expr' rather than 'A'?
>
> Are the comments swapped too?
>
> -Jordan
>

Oh, too much coffee I guess.
You are of course doubly correct.
Thanks for spotting that!
I'll put up a new version soon.

-Thomas

> > +      } else if (op_expr[1] && op_expr[1]->operation ==
ir_binop_logic_and &&
> > +                    (op_expr[1]->operands[0]->equals(op_expr[0]) ||
> > +                     op_expr[1]->operands[1]->equals(op_expr[0]))) {
> > +         /* (A && B) || A  or (B && A) || A */
> >
> > +         return ir->operands[1];
> >        }
> >        break;
> >
> > --
> > 2.0.0
> >
> > _______________________________________________
> > mesa-dev mailing list
> > mesa-dev at lists.freedesktop.org
> > http://lists.freedesktop.org/mailman/listinfo/mesa-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20140711/01951665/attachment.html>