Re: [Mesa-dev] [PATCH] nir: lower fexp2(fmul(flog2(a), 2)) to fmul(a, a)

[Samuel Pitoiset]

On 02/06/2018 12:29 AM, Ian Romanick wrote:
Do you have any data from shader-db for this change (and the other patch)?

We didn't see any changes for this patch on ivy bridge, but it helps 
when fpow is lowered.

shader-db results for the other patch https://hastebin.com/gajajodebo.hs 
on ivy bridge, seems good overall.


On 02/05/2018 06:08 AM, Samuel Pitoiset wrote:
Similar for the 4 case.

Suggested by Bas.

Signed-off-by: Samuel Pitoiset <samuel.pitoi...@gmail.com>
---
  src/compiler/nir/nir_opt_algebraic.py | 2 ++
  1 file changed, 2 insertions(+)

diff --git a/src/compiler/nir/nir_opt_algebraic.py 
b/src/compiler/nir/nir_opt_algebraic.py
index 6dc19d9b12..b30d1df199 100644
--- a/src/compiler/nir/nir_opt_algebraic.py
+++ b/src/compiler/nir/nir_opt_algebraic.py
@@ -321,6 +321,8 @@ optimizations = [
     (('~fexp2', ('fmul', ('flog2', a), b)), ('fpow', a, b), 
'!options->lower_fpow'), # 2^(lg2(a)*b) = a^b
     (('~fexp2', ('fadd', ('fmul', ('flog2', a), b), ('fmul', ('flog2', c), 
d))),
      ('~fmul', ('fpow', a, b), ('fpow', c, d)), '!options->lower_fpow'), # 
2^(lg2(a) * b + lg2(c) + d) = a^b * c^d
+   (('~fexp2', ('fmul', ('flog2', a), 2.0)), ('fmul', a, a)),
+   (('~fexp2', ('fmul', ('flog2', a), 4.0)), ('fmul', ('fmul', a, a), ('fmul', 
a, a))),
     (('~fpow', a, 1.0), a),
     (('~fpow', a, 2.0), ('fmul', a, a)),
     (('~fpow', a, 4.0), ('fmul', ('fmul', a, a), ('fmul', a, a))),


_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev