nir/algebraic: add option to lower fall_equalN/fany_nequalN

Add generic lowerings for fall_equalN/fany_nequalN. These should be optimal
for vec4 backends that doesn't have any special instructions for it, as
long as they support saturate.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Reviewed-by: Matt Turner <mattst88@gmail.com>

diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
index f41302676d1664d98b23d08913067d4321ea6248..b011f868bc2c37557e0c2d29cbe9fef143a4c0d7 100644 (file)
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -2537,6 +2537,9 @@ typedef struct nir_shader_compiler_options {
    /* lower {slt,sge,seq,sne} to {flt,fge,feq,fne} + b2f: */
    bool lower_scmp;
 
+   /* lower fall_equalN/fany_nequalN (ex:fany_nequal4 to sne+fdot4+fsat) */
+   bool lower_vector_cmp;
+
    /** enables rules to lower idiv by power-of-two: */
    bool lower_idiv;
 

diff --git a/src/compiler/nir/nir_opt_algebraic.py b/src/compiler/nir/nir_opt_algebraic.py
index 1ac1e6c64c0453fe4e6a4bdffd3b18454797cd39..e4f4a652c1914e0e41c9d1a41b63308b5d31b9ae 100644 (file)
--- a/src/compiler/nir/nir_opt_algebraic.py
+++ b/src/compiler/nir/nir_opt_algebraic.py
@@ -568,6 +568,12 @@ optimizations = [
    (('sge', a, b), ('b2f', ('fge', a, b)), 'options->lower_scmp'),
    (('seq', a, b), ('b2f', ('feq', a, b)), 'options->lower_scmp'),
    (('sne', a, b), ('b2f', ('fne', a, b)), 'options->lower_scmp'),
+   (('fall_equal2', a, b), ('fmin', ('seq', 'a.x', 'b.x'), ('seq', 'a.y', 'b.y')), 'options->lower_vector_cmp'),
+   (('fall_equal3', a, b), ('seq', ('fany_nequal3', a, b), 0.0), 'options->lower_vector_cmp'),
+   (('fall_equal4', a, b), ('seq', ('fany_nequal4', a, b), 0.0), 'options->lower_vector_cmp'),
+   (('fany_nequal2', a, b), ('fmax', ('sne', 'a.x', 'b.x'), ('sne', 'a.y', 'b.y')), 'options->lower_vector_cmp'),
+   (('fany_nequal3', a, b), ('fsat', ('fdot3', ('sne', a, b), ('sne', a, b))), 'options->lower_vector_cmp'),
+   (('fany_nequal4', a, b), ('fsat', ('fdot4', ('sne', a, b), ('sne', a, b))), 'options->lower_vector_cmp'),
    (('fne', ('fneg', a), a), ('fne', a, 0.0)),
    (('feq', ('fneg', a), a), ('feq', a, 0.0)),
    # Emulating booleans