nir: Add relaxed 24bit opcodes

These are equivalent to the 32bit opcodes if there are no more efficient
24bit opcodes available, but inputs are guaranteed to already be 24bit,
so the 24bit opcodes can be used instead if they exist and are efficient.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10549>
This commit is contained in:
Jesse Natalie 2021-04-30 11:46:09 -07:00 committed by Marge Bot
parent 9410eb7e39
commit d7ca0319d7
2 changed files with 13 additions and 0 deletions

View file

@ -1258,6 +1258,11 @@ triop("umad24", tuint32, _2src_commutative,
binop("umul24", tint32, _2src_commutative + associative,
"(((uint32_t)src0 << 8) >> 8) * (((uint32_t)src1 << 8) >> 8)")
# relaxed versions of the above, which assume input is in the 24bit range (no clamping)
binop("imul24_relaxed", tint32, _2src_commutative + associative, "src0 * src1")
triop("umad24_relaxed", tuint32, _2src_commutative, "src0 * src1 + src2")
binop("umul24_relaxed", tuint32, _2src_commutative + associative, "src0 * src1")
unop_convert("fisnormal", tbool1, tfloat, "isnormal(src0)")
unop_convert("fisfinite", tbool1, tfloat, "isfinite(src0)")

View file

@ -1653,6 +1653,14 @@ optimizations.extend([
('iadd', ('imul', ('iand', a, 0xffffff), ('iand', b, 0xffffff)), c),
'!options->has_umad24'),
# Relaxed 24bit ops
(('imul24_relaxed', a, b), ('imul24', a, b), 'options->has_imul24'),
(('imul24_relaxed', a, b), ('imul', a, b), '!options->has_imul24'),
(('umad24_relaxed', a, b, c), ('umad24', a, b, c), 'options->has_umad24'),
(('umad24_relaxed', a, b, c), ('iadd', ('umul24_relaxed', a, b), c), '!options->has_umad24'),
(('umul24_relaxed', a, b), ('umul24', a, b), 'options->has_umul24'),
(('umul24_relaxed', a, b), ('imul', a, b), '!options->has_umul24'),
(('imad24_ir3', a, b, 0), ('imul24', a, b)),
(('imad24_ir3', a, 0, c), (c)),
(('imad24_ir3', a, 1, c), ('iadd', a, c)),