[Mesa-dev] [PATCH 1/2] nir: Add some missing "optimization undo" patterns
Timothy Arceri
tarceri at itsqueeze.com
Fri Jun 1 00:38:05 UTC 2018
On 01/06/18 10:21, Ian Romanick wrote:
> From: Ian Romanick <ian.d.romanick at intel.com>
>
> d8d18516b0a and 03fb13f6467 added some patterns to undo conversions like
>
> (('ior', ('flt', a, b), ('flt', a, c)), ('flt', a, ('fmax', b, c)))
Reviewed-by: Timothy Arceri <tarceri at itsqueeze.com>
However can you explain where these conversions happen? Why do we not
get stuck in a loop converting back an forth?
>
> If further optimization cause some of the operands to either be the same
> or be constants, undoing the transformation can lead to further savings.
>
> I don't know why these patterns were not added in those patches. I did
> not check to see which specific patterns actually helped. I just added
> all of them for symmetry. This prevents some loop unrolling regressions
> Plane Shift caused by Samuel's "nir: implement the GLSL equivalent of if
> simplication in nir_opt_if" patch.
>
> Skylake and Broadwell had similar results. (Skylake shown)
> total instructions in shared programs: 14369768 -> 14369557 (<.01%)
> instructions in affected programs: 44076 -> 43865 (-0.48%)
> helped: 141
> HURT: 0
> helped stats (abs) min: 1 max: 5 x̄: 1.50 x̃: 1
> helped stats (rel) min: 0.07% max: 1.52% x̄: 0.66% x̃: 0.60%
> 95% mean confidence interval for instructions value: -1.67 -1.32
> 95% mean confidence interval for instructions %-change: -0.72% -0.59%
> Instructions are helped.
>
> total cycles in shared programs: 532430629 -> 532425772 (<.01%)
> cycles in affected programs: 1170832 -> 1165975 (-0.41%)
> helped: 101
> HURT: 5
> helped stats (abs) min: 1 max: 160 x̄: 48.54 x̃: 32
> helped stats (rel) min: <.01% max: 8.49% x̄: 2.76% x̃: 2.03%
> HURT stats (abs) min: 2 max: 22 x̄: 9.20 x̃: 4
> HURT stats (rel) min: <.01% max: 0.05% x̄: 0.02% x̃: <.01%
> 95% mean confidence interval for cycles value: -53.64 -38.00
> 95% mean confidence interval for cycles %-change: -3.06% -2.20%
> Cycles are helped.
>
> Signed-off-by: Ian Romanick <ian.d.romanick at intel.com>
> Cc: Samuel Pitoiset <samuel.pitoiset at gmail.com>
> Cc: Timothy Arceri <tarceri at itsqueeze.com>
> ---
> src/compiler/nir/nir_opt_algebraic.py | 20 +++++++++++++++++++-
> 1 file changed, 19 insertions(+), 1 deletion(-)
>
> diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
> index ba788f221a3..f153570105b 100644
> --- a/src/compiler/nir/nir_opt_algebraic.py
> +++ b/src/compiler/nir/nir_opt_algebraic.py
> @@ -292,6 +292,8 @@ optimizations = [
> (('~fge', ('fmax', a, b), a), True),
> (('~flt', a, ('fmin', b, a)), False),
> (('~flt', ('fmax', a, b), a), False),
> + (('~fge', a, ('fmax', b, a)), ('fge', a, b)),
> + (('~fge', ('fmin', a, b), a), ('fge', b, a)),
>
> (('ilt', a, ('imax', b, a)), ('ilt', a, b)),
> (('ilt', ('imin', a, b), a), ('ilt', b, a)),
> @@ -301,7 +303,23 @@ optimizations = [
> (('ult', ('umin', a, b), a), ('ult', b, a)),
> (('uge', a, ('umin', b, a)), True),
> (('uge', ('umax', a, b), a), True),
> -
> + (('ilt', a, ('imin', b, a)), False),
> + (('ilt', ('imax', a, b), a), False),
> + (('ige', a, ('imax', b, a)), ('ige', a, b)),
> + (('ige', ('imin', a, b), a), ('ige', b, a)),
> + (('ult', a, ('umin', b, a)), False),
> + (('ult', ('umax', a, b), a), False),
> + (('uge', a, ('umax', b, a)), ('uge', a, b)),
> + (('uge', ('umin', a, b), a), ('uge', b, a)),
> +
> + (('ilt', '#a', ('imax', '#b', c)), ('ior', ('ilt', a, b), ('ilt', a, c))),
> + (('ilt', ('imin', '#a', b), '#c'), ('ior', ('ilt', a, c), ('ilt', b, c))),
> + (('ige', '#a', ('imin', '#b', c)), ('ior', ('ige', a, b), ('ige', a, c))),
> + (('ige', ('imax', '#a', b), '#c'), ('ior', ('ige', a, c), ('ige', b, c))),
> + (('ult', '#a', ('umax', '#b', c)), ('ior', ('ult', a, b), ('ult', a, c))),
> + (('ult', ('umin', '#a', b), '#c'), ('ior', ('ult', a, c), ('ult', b, c))),
> + (('uge', '#a', ('umin', '#b', c)), ('ior', ('uge', a, b), ('uge', a, c))),
> + (('uge', ('umax', '#a', b), '#c'), ('ior', ('uge', a, c), ('uge', b, c))),
> (('ilt', '#a', ('imin', '#b', c)), ('iand', ('ilt', a, b), ('ilt', a, c))),
> (('ilt', ('imax', '#a', b), '#c'), ('iand', ('ilt', a, c), ('ilt', b, c))),
> (('ige', '#a', ('imax', '#b', c)), ('iand', ('ige', a, b), ('ige', a, c))),
>
More information about the mesa-dev
mailing list