# Authors:
# Jason Ekstrand (jason@jlekstrand.net)
+from __future__ import print_function
+
+from collections import OrderedDict
import nir_algebraic
+from nir_opcodes import type_sizes
+import itertools
# Convenience variables
a = 'a'
b = 'b'
c = 'c'
d = 'd'
+e = 'e'
# Written in the form (<search>, <replace>) where <search> is an expression
# and <replace> is either an expression or a value. An expression is
#
# All expression types can have a bit-size specified. For opcodes, this
# looks like "op@32", for variables it is "a@32" or "a@uint32" to specify a
-# type and size, and for literals, you can write "2.0@32". In the search half
-# of the expression this indicates that it should only match that particular
-# bit-size. In the replace half of the expression this indicates that the
-# constructed value should have that bit-size.
+# type and size. In the search half of the expression this indicates that it
+# should only match that particular bit-size. In the replace half of the
+# expression this indicates that the constructed value should have that
+# bit-size.
optimizations = [
(('imul', a, '#b@32(is_pos_power_of_two)'), ('ishl', a, ('find_lsb', b))),
(('imul', a, '#b@32(is_neg_power_of_two)'), ('ineg', ('ishl', a, ('find_lsb', ('iabs', b))))),
+ (('unpack_64_2x32_split_x', ('imul_2x32_64(is_used_once)', a, b)), ('imul', a, b)),
+ (('unpack_64_2x32_split_x', ('umul_2x32_64(is_used_once)', a, b)), ('imul', a, b)),
+ (('imul_2x32_64', a, b), ('pack_64_2x32_split', ('imul', a, b), ('imul_high', a, b)), 'options->lower_mul_2x32_64'),
+ (('umul_2x32_64', a, b), ('pack_64_2x32_split', ('imul', a, b), ('umul_high', a, b)), 'options->lower_mul_2x32_64'),
(('udiv', a, 1), a),
(('idiv', a, 1), a),
(('umod', a, 1), 0),
(('ineg', ('ineg', a)), a),
(('fabs', ('fabs', a)), ('fabs', a)),
(('fabs', ('fneg', a)), ('fabs', a)),
- (('fabs', ('u2f32', a)), ('u2f32', a)),
+ (('fabs', ('u2f', a)), ('u2f', a)),
(('iabs', ('iabs', a)), ('iabs', a)),
(('iabs', ('ineg', a)), ('iabs', a)),
+ (('f2b', ('fneg', a)), ('f2b', a)),
+ (('i2b', ('ineg', a)), ('i2b', a)),
(('~fadd', a, 0.0), a),
(('iadd', a, 0), a),
(('usadd_4x8', a, 0), a),
(('imul', a, 1), a),
(('fmul', a, -1.0), ('fneg', a)),
(('imul', a, -1), ('ineg', a)),
+ # If a < 0: fsign(a)*a*a => -1*a*a => -a*a => abs(a)*a
+ # If a > 0: fsign(a)*a*a => 1*a*a => a*a => abs(a)*a
+ # If a == 0: fsign(a)*a*a => 0*0*0 => abs(0)*0
+ (('fmul', ('fsign', a), ('fmul', a, a)), ('fmul', ('fabs', a), a)),
+ (('fmul', ('fmul', ('fsign', a), a), a), ('fmul', ('fabs', a), a)),
(('~ffma', 0.0, a, b), b),
(('~ffma', a, 0.0, b), b),
(('~ffma', a, b, 0.0), ('fmul', a, b)),
(('~flrp', a, b, 1.0), b),
(('~flrp', a, a, b), a),
(('~flrp', 0.0, a, b), ('fmul', a, b)),
- (('~flrp', a, b, ('b2f', c)), ('bcsel', c, b, a), 'options->lower_flrp32'),
+
+ # flrp(a, a + b, c) => a + flrp(0, b, c) => a + (b * c)
+ (('~flrp', a, ('fadd(is_used_once)', a, b), c), ('fadd', ('fmul', b, c), a)),
+ (('~flrp@32', a, ('fadd', a, b), c), ('fadd', ('fmul', b, c), a), 'options->lower_flrp32'),
+ (('~flrp@64', a, ('fadd', a, b), c), ('fadd', ('fmul', b, c), a), 'options->lower_flrp64'),
+
+ (('~flrp@32', ('fadd', a, b), ('fadd', a, c), d), ('fadd', ('flrp', b, c, d), a), 'options->lower_flrp32'),
+ (('~flrp@64', ('fadd', a, b), ('fadd', a, c), d), ('fadd', ('flrp', b, c, d), a), 'options->lower_flrp64'),
+
+ (('~flrp@32', a, ('fmul(is_used_once)', a, b), c), ('fmul', ('flrp', 1.0, b, c), a), 'options->lower_flrp32'),
+ (('~flrp@64', a, ('fmul(is_used_once)', a, b), c), ('fmul', ('flrp', 1.0, b, c), a), 'options->lower_flrp64'),
+
+ (('~flrp', ('fmul(is_used_once)', a, b), ('fmul(is_used_once)', a, c), d), ('fmul', ('flrp', b, c, d), a)),
+
+ (('~flrp', a, b, ('b2f', 'c@1')), ('bcsel', c, b, a), 'options->lower_flrp32'),
(('~flrp', a, 0.0, c), ('fadd', ('fmul', ('fneg', a), c), a)),
- (('flrp@32', a, b, c), ('fadd', ('fmul', c, ('fsub', b, a)), a), 'options->lower_flrp32'),
- (('flrp@64', a, b, c), ('fadd', ('fmul', c, ('fsub', b, a)), a), 'options->lower_flrp64'),
+ (('ftrunc', a), ('bcsel', ('flt', a, 0.0), ('fneg', ('ffloor', ('fabs', a))), ('ffloor', ('fabs', a))), 'options->lower_ftrunc'),
+ (('ffloor', a), ('fsub', a, ('ffract', a)), 'options->lower_ffloor'),
+ (('fadd', a, ('fneg', ('ffract', a))), ('ffloor', a), '!options->lower_ffloor'),
(('ffract', a), ('fsub', a, ('ffloor', a)), 'options->lower_ffract'),
- (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg', ('b2f', c)))), ('fmul', b, ('b2f', c))), ('bcsel', c, b, a), 'options->lower_flrp32'),
+ (('fceil', a), ('fneg', ('ffloor', ('fneg', a))), 'options->lower_fceil'),
+ (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg', ('b2f', 'c@1')))), ('fmul', b, ('b2f', c))), ('bcsel', c, b, a), 'options->lower_flrp32'),
(('~fadd@32', ('fmul', a, ('fadd', 1.0, ('fneg', c ))), ('fmul', b, c )), ('flrp', a, b, c), '!options->lower_flrp32'),
(('~fadd@64', ('fmul', a, ('fadd', 1.0, ('fneg', c ))), ('fmul', b, c )), ('flrp', a, b, c), '!options->lower_flrp64'),
- (('~fadd', a, ('fmul', ('b2f', c), ('fadd', b, ('fneg', a)))), ('bcsel', c, b, a), 'options->lower_flrp32'),
+ (('~fadd', a, ('fmul', ('b2f', 'c@1'), ('fadd', b, ('fneg', a)))), ('bcsel', c, b, a), 'options->lower_flrp32'),
(('~fadd@32', a, ('fmul', c , ('fadd', b, ('fneg', a)))), ('flrp', a, b, c), '!options->lower_flrp32'),
(('~fadd@64', a, ('fmul', c , ('fadd', b, ('fneg', a)))), ('flrp', a, b, c), '!options->lower_flrp64'),
(('ffma', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma'),
(('~fadd', ('fmul', a, b), c), ('ffma', a, b, c), 'options->fuse_ffma'),
+ (('~fmul', ('fadd', ('iand', ('ineg', ('b2i32', 'a@bool')), ('fmul', b, c)), '#d'), '#e'),
+ ('bcsel', a, ('fmul', ('fadd', ('fmul', b, c), d), e), ('fmul', d, e))),
+
+ (('fdot4', ('vec4', a, b, c, 1.0), d), ('fdph', ('vec3', a, b, c), d)),
+ (('fdot4', ('vec4', a, 0.0, 0.0, 0.0), b), ('fmul', a, b)),
+ (('fdot4', ('vec4', a, b, 0.0, 0.0), c), ('fdot2', ('vec2', a, b), c)),
+ (('fdot4', ('vec4', a, b, c, 0.0), d), ('fdot3', ('vec3', a, b, c), d)),
+
+ (('fdot3', ('vec3', a, 0.0, 0.0), b), ('fmul', a, b)),
+ (('fdot3', ('vec3', a, b, 0.0), c), ('fdot2', ('vec2', a, b), c)),
+
# (a * #b + #c) << #d
# ((a * #b) << #d) + (#c << #d)
# (a * (#b << #d)) + (#c << #d)
(('~inot', ('feq', a, b)), ('fne', a, b)),
(('~inot', ('fne', a, b)), ('feq', a, b)),
(('inot', ('ilt', a, b)), ('ige', a, b)),
+ (('inot', ('ult', a, b)), ('uge', a, b)),
(('inot', ('ige', a, b)), ('ilt', a, b)),
+ (('inot', ('uge', a, b)), ('ult', a, b)),
(('inot', ('ieq', a, b)), ('ine', a, b)),
(('inot', ('ine', a, b)), ('ieq', a, b)),
# b2f(a) <= 0.0
# b2f(a) == 0.0 because b2f(a) can only be 0 or 1
# inot(a)
- (('fge', 0.0, ('b2f', a)), ('inot', a)),
-
- (('fge', ('fneg', ('b2f', a)), 0.0), ('inot', a)),
+ (('fge', 0.0, ('b2f', 'a@1')), ('inot', a)),
+
+ (('fge', ('fneg', ('b2f', 'a@1')), 0.0), ('inot', a)),
+
+ (('fne', ('fadd', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('ior', a, b)),
+ (('fne', ('fmax', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('ior', a, b)),
+ (('fne', ('bcsel', a, 1.0, ('b2f', 'b@1')) , 0.0), ('ior', a, b)),
+ (('fne', ('b2f', 'a@1'), ('fneg', ('b2f', 'b@1'))), ('ior', a, b)),
+ (('fne', ('fmul', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('iand', a, b)),
+ (('fne', ('fmin', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('iand', a, b)),
+ (('fne', ('bcsel', a, ('b2f', 'b@1'), 0.0) , 0.0), ('iand', a, b)),
+ (('fne', ('fadd', ('b2f', 'a@1'), ('fneg', ('b2f', 'b@1'))), 0.0), ('ixor', a, b)),
+ (('fne', ('b2f', 'a@1') , ('b2f', 'b@1') ), ('ixor', a, b)),
+ (('fne', ('fneg', ('b2f', 'a@1')), ('fneg', ('b2f', 'b@1'))), ('ixor', a, b)),
+ (('feq', ('fadd', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('inot', ('ior', a, b))),
+ (('feq', ('fmax', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('inot', ('ior', a, b))),
+ (('feq', ('bcsel', a, 1.0, ('b2f', 'b@1')) , 0.0), ('inot', ('ior', a, b))),
+ (('feq', ('b2f', 'a@1'), ('fneg', ('b2f', 'b@1'))), ('inot', ('ior', a, b))),
+ (('feq', ('fmul', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('inot', ('iand', a, b))),
+ (('feq', ('fmin', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('inot', ('iand', a, b))),
+ (('feq', ('bcsel', a, ('b2f', 'b@1'), 0.0) , 0.0), ('inot', ('iand', a, b))),
+ (('feq', ('fadd', ('b2f', 'a@1'), ('fneg', ('b2f', 'b@1'))), 0.0), ('ieq', a, b)),
+ (('feq', ('b2f', 'a@1') , ('b2f', 'b@1') ), ('ieq', a, b)),
+ (('feq', ('fneg', ('b2f', 'a@1')), ('fneg', ('b2f', 'b@1'))), ('ieq', a, b)),
+
+ # -(b2f(a) + b2f(b)) < 0
+ # 0 < b2f(a) + b2f(b)
+ # 0 != b2f(a) + b2f(b) b2f must be 0 or 1, so the sum is non-negative
+ # a || b
+ (('flt', ('fneg', ('fadd', ('b2f', 'a@1'), ('b2f', 'b@1'))), 0.0), ('ior', a, b)),
+ (('flt', 0.0, ('fadd', ('b2f', 'a@1'), ('b2f', 'b@1'))), ('ior', a, b)),
+
+ # -(b2f(a) + b2f(b)) >= 0
+ # 0 >= b2f(a) + b2f(b)
+ # 0 == b2f(a) + b2f(b) b2f must be 0 or 1, so the sum is non-negative
+ # !(a || b)
+ (('fge', ('fneg', ('fadd', ('b2f', 'a@1'), ('b2f', 'b@1'))), 0.0), ('inot', ('ior', a, b))),
+ (('fge', 0.0, ('fadd', ('b2f', 'a@1'), ('b2f', 'b@1'))), ('inot', ('ior', a, b))),
+
+ (('flt', a, ('fneg', a)), ('flt', a, 0.0)),
+ (('fge', a, ('fneg', a)), ('fge', a, 0.0)),
+
+ # Some optimizations (below) convert things like (a < b || c < b) into
+ # (min(a, c) < b). However, this interfers with the previous optimizations
+ # that try to remove comparisons with negated sums of b2f. This just
+ # breaks that apart.
+ (('flt', ('fmin', c, ('fneg', ('fadd', ('b2f', 'a@1'), ('b2f', 'b@1')))), 0.0),
+ ('ior', ('flt', c, 0.0), ('ior', a, b))),
+
+ (('~flt', ('fadd', a, b), a), ('flt', b, 0.0)),
+ (('~fge', ('fadd', a, b), a), ('fge', b, 0.0)),
+ (('~feq', ('fadd', a, b), a), ('feq', b, 0.0)),
+ (('~fne', ('fadd', a, b), a), ('fne', b, 0.0)),
+
+ # Cannot remove the addition from ilt or ige due to overflow.
+ (('ieq', ('iadd', a, b), a), ('ieq', b, 0)),
+ (('ine', ('iadd', a, b), a), ('ine', b, 0)),
# fmin(-b2f(a), b) >= 0.0
# -b2f(a) >= 0.0 && b >= 0.0
# The fge in the second replacement is not a typo. I leave the proof that
# "fmin(-b2f(a), b) >= 0 <=> fmin(-b2f(a), b) == 0" as an exercise for the
# reader.
- (('fge', ('fmin', ('fneg', ('b2f', a)), b), 0.0), ('iand', ('inot', a), ('fge', b, 0.0))),
- (('feq', ('fmin', ('fneg', ('b2f', a)), b), 0.0), ('iand', ('inot', a), ('fge', b, 0.0))),
+ (('fge', ('fmin', ('fneg', ('b2f', 'a@1')), 'b@1'), 0.0), ('iand', ('inot', a), ('fge', b, 0.0))),
+ (('feq', ('fmin', ('fneg', ('b2f', 'a@1')), 'b@1'), 0.0), ('iand', ('inot', a), ('fge', b, 0.0))),
+
+ (('feq', ('b2f', 'a@1'), 0.0), ('inot', a)),
+ (('fne', ('b2f', 'a@1'), 0.0), a),
+ (('ieq', ('b2i', 'a@1'), 0), ('inot', a)),
+ (('ine', ('b2i', 'a@1'), 0), a),
+
+ (('fne', ('u2f', a), 0.0), ('ine', a, 0)),
+ (('feq', ('u2f', a), 0.0), ('ieq', a, 0)),
+ (('fge', ('u2f', a), 0.0), True),
+ (('fge', 0.0, ('u2f', a)), ('uge', 0, a)), # ieq instead?
+ (('flt', ('u2f', a), 0.0), False),
+ (('flt', 0.0, ('u2f', a)), ('ult', 0, a)), # ine instead?
+ (('fne', ('i2f', a), 0.0), ('ine', a, 0)),
+ (('feq', ('i2f', a), 0.0), ('ieq', a, 0)),
+ (('fge', ('i2f', a), 0.0), ('ige', a, 0)),
+ (('fge', 0.0, ('i2f', a)), ('ige', 0, a)),
+ (('flt', ('i2f', a), 0.0), ('ilt', a, 0)),
+ (('flt', 0.0, ('i2f', a)), ('ilt', 0, a)),
# 0.0 < fabs(a)
# fabs(a) > 0.0
# fabs(a) != 0.0 because fabs(a) must be >= 0
# a != 0.0
- (('flt', 0.0, ('fabs', a)), ('fne', a, 0.0)),
+ (('~flt', 0.0, ('fabs', a)), ('fne', a, 0.0)),
+
+ # -fabs(a) < 0.0
+ # fabs(a) > 0.0
+ (('~flt', ('fneg', ('fabs', a)), 0.0), ('fne', a, 0.0)),
+
+ # 0.0 >= fabs(a)
+ # 0.0 == fabs(a) because fabs(a) must be >= 0
+ # 0.0 == a
+ (('fge', 0.0, ('fabs', a)), ('feq', a, 0.0)),
- (('fmax', ('b2f(is_used_once)', a), ('b2f', b)), ('b2f', ('ior', a, b))),
- (('fmax', ('fneg(is_used_once)', ('b2f(is_used_once)', a)), ('fneg', ('b2f', b))), ('fneg', ('b2f', ('ior', a, b)))),
- (('fmin', ('b2f(is_used_once)', a), ('b2f', b)), ('b2f', ('iand', a, b))),
- (('fmin', ('fneg(is_used_once)', ('b2f(is_used_once)', a)), ('fneg', ('b2f', b))), ('fneg', ('b2f', ('iand', a, b)))),
+ # -fabs(a) >= 0.0
+ # 0.0 >= fabs(a)
+ (('fge', ('fneg', ('fabs', a)), 0.0), ('feq', a, 0.0)),
+
+ (('fmax', ('b2f(is_used_once)', 'a@1'), ('b2f', 'b@1')), ('b2f', ('ior', a, b))),
+ (('fmax', ('fneg(is_used_once)', ('b2f(is_used_once)', 'a@1')), ('fneg', ('b2f', 'b@1'))), ('fneg', ('b2f', ('ior', a, b)))),
+ (('fmin', ('b2f(is_used_once)', 'a@1'), ('b2f', 'b@1')), ('b2f', ('iand', a, b))),
+ (('fmin', ('fneg(is_used_once)', ('b2f(is_used_once)', 'a@1')), ('fneg', ('b2f', 'b@1'))), ('fneg', ('b2f', ('iand', a, b)))),
+
+ # fmin(b2f(a), b)
+ # bcsel(a, fmin(b2f(a), b), fmin(b2f(a), b))
+ # bcsel(a, fmin(b2f(True), b), fmin(b2f(False), b))
+ # bcsel(a, fmin(1.0, b), fmin(0.0, b))
+ #
+ # Since b is a constant, constant folding will eliminate the fmin and the
+ # fmax. If b is > 1.0, the bcsel will be replaced with a b2f.
+ (('fmin', ('b2f', 'a@1'), '#b'), ('bcsel', a, ('fmin', b, 1.0), ('fmin', b, 0.0))),
- # ignore this opt when the result is used by a bcsel or if so we can make
- # use of conditional modifiers on supported hardware.
- (('flt(is_not_used_by_conditional)', ('fadd(is_used_once)', a, ('fneg', b)), 0.0), ('flt', a, b)),
+ (('flt', ('fadd(is_used_once)', a, ('fneg', b)), 0.0), ('flt', a, b)),
(('fge', ('fneg', ('fabs', a)), 0.0), ('feq', a, 0.0)),
(('~bcsel', ('flt', b, a), b, a), ('fmin', a, b)),
(('~bcsel', ('flt', a, b), b, a), ('fmax', a, b)),
(('~bcsel', ('fge', a, b), b, a), ('fmin', a, b)),
(('~bcsel', ('fge', b, a), b, a), ('fmax', a, b)),
+ (('bcsel', ('i2b', a), b, c), ('bcsel', ('ine', a, 0), b, c)),
(('bcsel', ('inot', a), b, c), ('bcsel', a, c, b)),
(('bcsel', a, ('bcsel', a, b, c), d), ('bcsel', a, b, d)),
- (('bcsel', a, True, 'b@bool'), ('ior', a, b)),
+ (('bcsel', a, b, ('bcsel', a, c, d)), ('bcsel', a, b, d)),
+ (('bcsel', a, ('bcsel', b, c, d), ('bcsel(is_used_once)', b, c, 'e')), ('bcsel', b, c, ('bcsel', a, d, 'e'))),
+ (('bcsel', a, ('bcsel(is_used_once)', b, c, d), ('bcsel', b, c, 'e')), ('bcsel', b, c, ('bcsel', a, d, 'e'))),
+ (('bcsel', a, ('bcsel', b, c, d), ('bcsel(is_used_once)', b, 'e', d)), ('bcsel', b, ('bcsel', a, c, 'e'), d)),
+ (('bcsel', a, ('bcsel(is_used_once)', b, c, d), ('bcsel', b, 'e', d)), ('bcsel', b, ('bcsel', a, c, 'e'), d)),
+ (('bcsel', a, True, b), ('ior', a, b)),
+ (('bcsel', a, a, b), ('ior', a, b)),
+ (('bcsel', a, b, False), ('iand', a, b)),
+ (('bcsel', a, b, a), ('iand', a, b)),
(('fmin', a, a), a),
(('fmax', a, a), a),
(('imin', a, a), a),
(('imax', a, a), a),
(('umin', a, a), a),
(('umax', a, a), a),
+ (('fmax', ('fmax', a, b), b), ('fmax', a, b)),
+ (('umax', ('umax', a, b), b), ('umax', a, b)),
+ (('imax', ('imax', a, b), b), ('imax', a, b)),
+ (('fmin', ('fmin', a, b), b), ('fmin', a, b)),
+ (('umin', ('umin', a, b), b), ('umin', a, b)),
+ (('imin', ('imin', a, b), b), ('imin', a, b)),
+ (('fmax', a, ('fneg', a)), ('fabs', a)),
+ (('imax', a, ('ineg', a)), ('iabs', a)),
(('fmin', a, ('fneg', a)), ('fneg', ('fabs', a))),
(('imin', a, ('ineg', a)), ('ineg', ('iabs', a))),
(('fmin', a, ('fneg', ('fabs', a))), ('fneg', ('fabs', a))),
(('imax', a, ('ineg', a)), ('iabs', a)),
(('~fmin', ('fmax', a, 0.0), 1.0), ('fsat', a), '!options->lower_fsat'),
(('~fmax', ('fmin', a, 1.0), 0.0), ('fsat', a), '!options->lower_fsat'),
+ (('fsat', ('fsign', a)), ('b2f', ('flt', 0.0, a))),
+ (('fsat', ('b2f', a)), ('b2f', a)),
(('fsat', a), ('fmin', ('fmax', a, 0.0), 1.0), 'options->lower_fsat'),
(('fsat', ('fsat', a)), ('fsat', a)),
(('fmin', ('fmax', ('fmin', ('fmax', a, b), c), b), c), ('fmin', ('fmax', a, b), c)),
(('iand', ('uge(is_used_once)', a, b), ('uge', a, c)), ('uge', a, ('umax', b, c))),
(('iand', ('uge(is_used_once)', a, c), ('uge', b, c)), ('uge', ('umin', a, b), c)),
+ # Common pattern like 'if (i == 0 || i == 1 || ...)'
+ (('ior', ('ieq', a, 0), ('ieq', a, 1)), ('uge', 1, a)),
+ (('ior', ('uge', 1, a), ('ieq', a, 2)), ('uge', 2, a)),
+ (('ior', ('uge', 2, a), ('ieq', a, 3)), ('uge', 3, a)),
+
+ # The (i2f32, ...) part is an open-coded fsign. When that is combined with
+ # the bcsel, it's basically copysign(1.0, a). There is no copysign in NIR,
+ # so emit an open-coded version of that.
+ (('bcsel@32', ('feq', a, 0.0), 1.0, ('i2f32', ('iadd', ('b2i32', ('flt', 0.0, 'a@32')), ('ineg', ('b2i32', ('flt', 'a@32', 0.0)))))),
+ ('ior', 0x3f800000, ('iand', a, 0x80000000))),
+
+ (('ior', a, ('ieq', a, False)), True),
+ (('ior', a, ('inot', a)), -1),
+
+ (('ine', ('ineg', ('b2i32', 'a@1')), ('ineg', ('b2i32', 'b@1'))), ('ine', a, b)),
+ (('b2i32', ('ine', 'a@1', 'b@1')), ('b2i32', ('ixor', a, b))),
+
+ (('iand', ('ieq', 'a@32', 0), ('ieq', 'b@32', 0)), ('ieq', ('ior', 'a@32', 'b@32'), 0)),
+
# These patterns can result when (a < b || a < c) => (a < min(b, c))
# transformations occur before constant propagation and loop-unrolling.
(('~flt', a, ('fmax', b, a)), ('flt', a, b)),
(('~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)),
(('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))),
(('uge', '#a', ('umax', '#b', c)), ('iand', ('uge', a, b), ('uge', a, c))),
(('uge', ('umin', '#a', b), '#c'), ('iand', ('uge', a, c), ('uge', b, c))),
+ # Thanks to sign extension, the ishr(a, b) is negative if and only if a is
+ # negative.
+ (('bcsel', ('ilt', a, 0), ('ineg', ('ishr', a, b)), ('ishr', a, b)),
+ ('iabs', ('ishr', a, b))),
+ (('iabs', ('ishr', ('iabs', a), b)), ('ishr', ('iabs', a), b)),
+
(('fabs', ('slt', a, b)), ('slt', a, b)),
(('fabs', ('sge', a, b)), ('sge', a, b)),
(('fabs', ('seq', a, b)), ('seq', a, b)),
(('fne', ('fneg', a), a), ('fne', a, 0.0)),
(('feq', ('fneg', a), a), ('feq', a, 0.0)),
# Emulating booleans
- (('imul', ('b2i', a), ('b2i', b)), ('b2i', ('iand', a, b))),
- (('fmul', ('b2f', a), ('b2f', b)), ('b2f', ('iand', a, b))),
- (('fsat', ('fadd', ('b2f', a), ('b2f', b))), ('b2f', ('ior', a, b))),
- (('iand', 'a@bool', 1.0), ('b2f', a)),
+ (('imul', ('b2i', 'a@1'), ('b2i', 'b@1')), ('b2i', ('iand', a, b))),
+ (('fmul', ('b2f', 'a@1'), ('b2f', 'b@1')), ('b2f', ('iand', a, b))),
+ (('fsat', ('fadd', ('b2f', 'a@1'), ('b2f', 'b@1'))), ('b2f', ('ior', a, b))),
+ (('iand', 'a@bool32', 1.0), ('b2f', a)),
# True/False are ~0 and 0 in NIR. b2i of True is 1, and -1 is ~0 (True).
- (('ineg', ('b2i@32', a)), a),
- (('flt', ('fneg', ('b2f', a)), 0), a), # Generated by TGSI KILL_IF.
- (('flt', ('fsub', 0.0, ('b2f', a)), 0), a), # Generated by TGSI KILL_IF.
+ (('ineg', ('b2i32', 'a@32')), a),
+ (('flt', ('fneg', ('b2f', 'a@1')), 0), a), # Generated by TGSI KILL_IF.
+ (('flt', ('fsub', 0.0, ('b2f', 'a@1')), 0), a), # Generated by TGSI KILL_IF.
# Comparison with the same args. Note that these are not done for
# the float versions because NaN always returns false on float
# inequalities.
(('ixor', a, a), 0),
(('ixor', a, 0), a),
(('inot', ('inot', a)), a),
+ (('ior', ('iand', a, b), b), b),
+ (('ior', ('ior', a, b), b), ('ior', a, b)),
+ (('iand', ('ior', a, b), b), b),
+ (('iand', ('iand', a, b), b), ('iand', a, b)),
# DeMorgan's Laws
(('iand', ('inot', a), ('inot', b)), ('inot', ('ior', a, b))),
(('ior', ('inot', a), ('inot', b)), ('inot', ('iand', a, b))),
(('fsqrt', a), ('frcp', ('frsq', a)), 'options->lower_fsqrt'),
(('~frcp', ('frsq', a)), ('fsqrt', a), '!options->lower_fsqrt'),
# Boolean simplifications
- (('ieq', 'a@bool', True), a),
- (('ine(is_not_used_by_if)', 'a@bool', True), ('inot', a)),
- (('ine', 'a@bool', False), a),
- (('ieq(is_not_used_by_if)', 'a@bool', False), ('inot', 'a')),
+ (('i2b32(is_used_by_if)', a), ('ine32', a, 0)),
+ (('i2b1(is_used_by_if)', a), ('ine', a, 0)),
+ (('ieq', a, True), a),
+ (('ine(is_not_used_by_if)', a, True), ('inot', a)),
+ (('ine', a, False), a),
+ (('ieq(is_not_used_by_if)', a, False), ('inot', 'a')),
(('bcsel', a, True, False), a),
(('bcsel', a, False, True), ('inot', a)),
(('bcsel@32', a, 1.0, 0.0), ('b2f', a)),
(('bcsel@32', a, -0.0, -1.0), ('fneg', ('b2f', ('inot', a)))),
(('bcsel', True, b, c), b),
(('bcsel', False, b, c), c),
+ (('bcsel', a, ('b2f(is_used_once)', 'b@32'), ('b2f', 'c@32')), ('b2f', ('bcsel', a, b, c))),
# The result of this should be hit by constant propagation and, in the
# next round of opt_algebraic, get picked up by one of the above two.
(('bcsel', '#a', b, c), ('bcsel', ('ine', 'a', 0), b, c)),
(('bcsel', a, b, b), b),
(('fcsel', a, b, b), b),
+ # D3D Boolean emulation
+ (('bcsel', a, -1, 0), ('ineg', ('b2i', 'a@1'))),
+ (('bcsel', a, 0, -1), ('ineg', ('b2i', ('inot', a)))),
+ (('iand', ('ineg', ('b2i', 'a@1')), ('ineg', ('b2i', 'b@1'))),
+ ('ineg', ('b2i', ('iand', a, b)))),
+ (('ior', ('ineg', ('b2i','a@1')), ('ineg', ('b2i', 'b@1'))),
+ ('ineg', ('b2i', ('ior', a, b)))),
+ (('ieq', ('ineg', ('b2i', 'a@1')), 0), ('inot', a)),
+ (('ieq', ('ineg', ('b2i', 'a@1')), -1), a),
+ (('ine', ('ineg', ('b2i', 'a@1')), 0), a),
+ (('ine', ('ineg', ('b2i', 'a@1')), -1), ('inot', a)),
+ (('iand', ('ineg', ('b2i', a)), 1.0), ('b2f', a)),
+
+ # SM5 32-bit shifts are defined to use the 5 least significant bits
+ (('ishl', 'a@32', ('iand', 31, b)), ('ishl', a, b)),
+ (('ishr', 'a@32', ('iand', 31, b)), ('ishr', a, b)),
+ (('ushr', 'a@32', ('iand', 31, b)), ('ushr', a, b)),
+
# Conversions
- (('i2b', ('b2i', a)), a),
- (('f2i32', ('ftrunc', a)), ('f2i32', a)),
- (('f2u32', ('ftrunc', a)), ('f2u32', a)),
+ (('i2b32', ('b2i', 'a@32')), a),
+ (('f2i', ('ftrunc', a)), ('f2i', a)),
+ (('f2u', ('ftrunc', a)), ('f2u', a)),
(('i2b', ('ineg', a)), ('i2b', a)),
(('i2b', ('iabs', a)), ('i2b', a)),
(('fabs', ('b2f', a)), ('b2f', a)),
(('iabs', ('b2i', a)), ('b2i', a)),
+ (('inot', ('f2b1', a)), ('feq', a, 0.0)),
+
+ # Ironically, mark these as imprecise because removing the conversions may
+ # preserve more precision than doing the conversions (e.g.,
+ # uint(float(0x81818181u)) == 0x81818200).
+ (('~f2i32', ('i2f', 'a@32')), a),
+ (('~f2i32', ('u2f', 'a@32')), a),
+ (('~f2u32', ('i2f', 'a@32')), a),
+ (('~f2u32', ('u2f', 'a@32')), a),
+
+ # Section 5.4.1 (Conversion and Scalar Constructors) of the GLSL 4.60 spec
+ # says:
+ #
+ # It is undefined to convert a negative floating-point value to an
+ # uint.
+ #
+ # Assuming that (uint)some_float behaves like (uint)(int)some_float allows
+ # some optimizations in the i965 backend to proceed.
+ (('ige', ('f2u', a), b), ('ige', ('f2i', a), b)),
+ (('ige', b, ('f2u', a)), ('ige', b, ('f2i', a))),
+ (('ilt', ('f2u', a), b), ('ilt', ('f2i', a), b)),
+ (('ilt', b, ('f2u', a)), ('ilt', b, ('f2i', a))),
# Packing and then unpacking does nothing
(('unpack_64_2x32_split_x', ('pack_64_2x32_split', a, b)), a),
('unpack_64_2x32_split_y', a)), a),
# Byte extraction
- (('ushr', a, 24), ('extract_u8', a, 3), '!options->lower_extract_byte'),
- (('iand', 0xff, ('ushr', a, 16)), ('extract_u8', a, 2), '!options->lower_extract_byte'),
- (('iand', 0xff, ('ushr', a, 8)), ('extract_u8', a, 1), '!options->lower_extract_byte'),
- (('iand', 0xff, a), ('extract_u8', a, 0), '!options->lower_extract_byte'),
+ (('ushr', 'a@16', 8), ('extract_u8', a, 1), '!options->lower_extract_byte'),
+ (('ushr', 'a@32', 24), ('extract_u8', a, 3), '!options->lower_extract_byte'),
+ (('ushr', 'a@64', 56), ('extract_u8', a, 7), '!options->lower_extract_byte'),
+ (('ishr', 'a@16', 8), ('extract_i8', a, 1), '!options->lower_extract_byte'),
+ (('ishr', 'a@32', 24), ('extract_i8', a, 3), '!options->lower_extract_byte'),
+ (('ishr', 'a@64', 56), ('extract_i8', a, 7), '!options->lower_extract_byte'),
+ (('iand', 0xff, a), ('extract_u8', a, 0), '!options->lower_extract_byte')
+]
+
+# After the ('extract_u8', a, 0) pattern, above, triggers, there will be
+# patterns like those below.
+for op in ('ushr', 'ishr'):
+ optimizations.extend([(('extract_u8', (op, 'a@16', 8), 0), ('extract_u8', a, 1))])
+ optimizations.extend([(('extract_u8', (op, 'a@32', 8 * i), 0), ('extract_u8', a, i)) for i in range(1, 4)])
+ optimizations.extend([(('extract_u8', (op, 'a@64', 8 * i), 0), ('extract_u8', a, i)) for i in range(1, 8)])
+optimizations.extend([(('extract_u8', ('extract_u16', a, 1), 0), ('extract_u8', a, 2))])
+
+# After the ('extract_[iu]8', a, 3) patterns, above, trigger, there will be
+# patterns like those below.
+for op in ('extract_u8', 'extract_i8'):
+ optimizations.extend([((op, ('ishl', 'a@16', 8), 1), (op, a, 0))])
+ optimizations.extend([((op, ('ishl', 'a@32', 24 - 8 * i), 3), (op, a, i)) for i in range(2, -1, -1)])
+ optimizations.extend([((op, ('ishl', 'a@64', 56 - 8 * i), 7), (op, a, i)) for i in range(6, -1, -1)])
+
+optimizations.extend([
# Word extraction
- (('ushr', a, 16), ('extract_u16', a, 1), '!options->lower_extract_word'),
+ (('ushr', ('ishl', 'a@32', 16), 16), ('extract_u16', a, 0), '!options->lower_extract_word'),
+ (('ushr', 'a@32', 16), ('extract_u16', a, 1), '!options->lower_extract_word'),
+ (('ishr', ('ishl', 'a@32', 16), 16), ('extract_i16', a, 0), '!options->lower_extract_word'),
+ (('ishr', 'a@32', 16), ('extract_i16', a, 1), '!options->lower_extract_word'),
(('iand', 0xffff, a), ('extract_u16', a, 0), '!options->lower_extract_word'),
# Subtracts
# Reassociate constants in add/mul chains so they can be folded together.
# For now, we mostly only handle cases where the constants are separated by
# a single non-constant. We could do better eventually.
- (('~fmul', '#a', ('fmul', b, '#c')), ('fmul', ('fmul', a, c), b)),
- (('imul', '#a', ('imul', b, '#c')), ('imul', ('imul', a, c), b)),
- (('~fadd', '#a', ('fadd', b, '#c')), ('fadd', ('fadd', a, c), b)),
- (('~fadd', '#a', ('fneg', ('fadd', b, '#c'))), ('fadd', ('fadd', a, ('fneg', c)), ('fneg', b))),
- (('iadd', '#a', ('iadd', b, '#c')), ('iadd', ('iadd', a, c), b)),
+ (('~fmul', '#a', ('fmul', 'b(is_not_const)', '#c')), ('fmul', ('fmul', a, c), b)),
+ (('imul', '#a', ('imul', 'b(is_not_const)', '#c')), ('imul', ('imul', a, c), b)),
+ (('~fadd', '#a', ('fadd', 'b(is_not_const)', '#c')), ('fadd', ('fadd', a, c), b)),
+ (('~fadd', '#a', ('fneg', ('fadd', 'b(is_not_const)', '#c'))), ('fadd', ('fadd', a, ('fneg', c)), ('fneg', b))),
+ (('iadd', '#a', ('iadd', 'b(is_not_const)', '#c')), ('iadd', ('iadd', a, c), b)),
# By definition...
(('bcsel', ('ige', ('find_lsb', a), 0), ('find_lsb', a), -1), ('find_lsb', a)),
(('bcsel', ('ine', a, -1), ('ifind_msb', a), -1), ('ifind_msb', a)),
# Misc. lowering
+ (('fmod@16', a, b), ('fsub', a, ('fmul', b, ('ffloor', ('fdiv', a, b)))), 'options->lower_fmod16'),
(('fmod@32', a, b), ('fsub', a, ('fmul', b, ('ffloor', ('fdiv', a, b)))), 'options->lower_fmod32'),
(('fmod@64', a, b), ('fsub', a, ('fmul', b, ('ffloor', ('fdiv', a, b)))), 'options->lower_fmod64'),
(('frem', a, b), ('fsub', a, ('fmul', b, ('ftrunc', ('fdiv', a, b)))), 'options->lower_fmod32'),
('bcsel', ('ilt', 31, 'bits'), 'insert',
('bfi', ('bfm', 'bits', 'offset'), 'insert', 'base')),
'options->lower_bitfield_insert'),
+ (('ihadd', a, b), ('iadd', ('iand', a, b), ('ishr', ('ixor', a, b), 1)), 'options->lower_hadd'),
+ (('uhadd', a, b), ('iadd', ('iand', a, b), ('ushr', ('ixor', a, b), 1)), 'options->lower_hadd'),
+ (('irhadd', a, b), ('isub', ('ior', a, b), ('ishr', ('ixor', a, b), 1)), 'options->lower_hadd'),
+ (('urhadd', a, b), ('isub', ('ior', a, b), ('ushr', ('ixor', a, b), 1)), 'options->lower_hadd'),
+ (('uadd_sat', a, b), ('bcsel', ('ult', ('iadd', a, b), a), -1, ('iadd', a, b)), 'options->lower_add_sat'),
+ (('usub_sat', a, b), ('bcsel', ('ult', a, b), 0, ('isub', a, b)), 'options->lower_add_sat'),
+
+ # Alternative lowering that doesn't rely on bfi.
+ (('bitfield_insert', 'base', 'insert', 'offset', 'bits'),
+ ('bcsel', ('ilt', 31, 'bits'),
+ 'insert',
+ ('ior',
+ ('iand', 'base', ('inot', ('bfm', 'bits', 'offset'))),
+ ('iand', ('ishl', 'insert', 'offset'), ('bfm', 'bits', 'offset')))),
+ 'options->lower_bitfield_insert_to_shifts'),
+
+ # bfm lowering -- note that the NIR opcode is undefined if either arg is 32.
+ (('bfm', 'bits', 'offset'),
+ ('ishl', ('isub', ('ishl', 1, 'bits'), 1), 'offset'),
+ 'options->lower_bfm'),
(('ibitfield_extract', 'value', 'offset', 'bits'),
('bcsel', ('ilt', 31, 'bits'), 'value',
('ubfe', 'value', 'offset', 'bits')),
'options->lower_bitfield_extract'),
+ (('ibitfield_extract', 'value', 'offset', 'bits'),
+ ('bcsel', ('ieq', 0, 'bits'),
+ 0,
+ ('ishr',
+ ('ishl', 'value', ('isub', ('isub', 32, 'bits'), 'offset')),
+ ('isub', 32, 'bits'))),
+ 'options->lower_bitfield_extract_to_shifts'),
+
+ (('ubitfield_extract', 'value', 'offset', 'bits'),
+ ('iand',
+ ('ushr', 'value', 'offset'),
+ ('bcsel', ('ieq', 'bits', 32),
+ 0xffffffff,
+ ('bfm', 'bits', 0))),
+ 'options->lower_bitfield_extract_to_shifts'),
+
+ (('ifind_msb', 'value'),
+ ('ufind_msb', ('bcsel', ('ilt', 'value', 0), ('inot', 'value'), 'value')),
+ 'options->lower_ifind_msb'),
+
+ (('find_lsb', 'value'),
+ ('ufind_msb', ('iand', 'value', ('ineg', 'value'))),
+ 'options->lower_find_lsb'),
+
(('extract_i8', a, 'b@32'),
('ishr', ('ishl', a, ('imul', ('isub', 3, b), 8)), 24),
'options->lower_extract_byte'),
(('unpack_unorm_2x16', 'v'),
('fdiv', ('u2f32', ('vec2', ('extract_u16', 'v', 0),
- ('extract_u16', 'v', 1))),
+ ('extract_u16', 'v', 1))),
65535.0),
'options->lower_unpack_unorm_2x16'),
(('unpack_unorm_4x8', 'v'),
('fdiv', ('u2f32', ('vec4', ('extract_u8', 'v', 0),
- ('extract_u8', 'v', 1),
- ('extract_u8', 'v', 2),
- ('extract_u8', 'v', 3))),
+ ('extract_u8', 'v', 1),
+ ('extract_u8', 'v', 2),
+ ('extract_u8', 'v', 3))),
255.0),
'options->lower_unpack_unorm_4x8'),
(('unpack_snorm_2x16', 'v'),
- ('fmin', 1.0, ('fmax', -1.0, ('fdiv', ('i2f32', ('vec2', ('extract_i16', 'v', 0),
- ('extract_i16', 'v', 1))),
+ ('fmin', 1.0, ('fmax', -1.0, ('fdiv', ('i2f', ('vec2', ('extract_i16', 'v', 0),
+ ('extract_i16', 'v', 1))),
32767.0))),
'options->lower_unpack_snorm_2x16'),
(('unpack_snorm_4x8', 'v'),
- ('fmin', 1.0, ('fmax', -1.0, ('fdiv', ('i2f32', ('vec4', ('extract_i8', 'v', 0),
- ('extract_i8', 'v', 1),
- ('extract_i8', 'v', 2),
- ('extract_i8', 'v', 3))),
+ ('fmin', 1.0, ('fmax', -1.0, ('fdiv', ('i2f', ('vec4', ('extract_i8', 'v', 0),
+ ('extract_i8', 'v', 1),
+ ('extract_i8', 'v', 2),
+ ('extract_i8', 'v', 3))),
127.0))),
'options->lower_unpack_snorm_4x8'),
-]
+
+ (('isign', a), ('imin', ('imax', a, -1), 1), 'options->lower_isign'),
+ (('fsign', a), ('fsub', ('b2f', ('flt', 0.0, a)), ('b2f', ('flt', a, 0.0))), 'options->lower_fsign'),
+])
+
+# bit_size dependent lowerings
+for bit_size in [8, 16, 32, 64]:
+ # convenience constants
+ intmax = (1 << (bit_size - 1)) - 1
+ intmin = 1 << (bit_size - 1)
+
+ optimizations += [
+ (('iadd_sat@' + str(bit_size), a, b),
+ ('bcsel', ('ige', b, 1), ('bcsel', ('ilt', ('iadd', a, b), a), intmax, ('iadd', a, b)),
+ ('bcsel', ('ilt', a, ('iadd', a, b)), intmin, ('iadd', a, b))), 'options->lower_add_sat'),
+ (('isub_sat@' + str(bit_size), a, b),
+ ('bcsel', ('ilt', b, 0), ('bcsel', ('ilt', ('isub', a, b), a), intmax, ('isub', a, b)),
+ ('bcsel', ('ilt', a, ('isub', a, b)), intmin, ('isub', a, b))), 'options->lower_add_sat'),
+ ]
+
+invert = OrderedDict([('feq', 'fne'), ('fne', 'feq'), ('fge', 'flt'), ('flt', 'fge')])
+
+for left, right in itertools.combinations_with_replacement(invert.keys(), 2):
+ optimizations.append((('inot', ('ior(is_used_once)', (left, a, b), (right, c, d))),
+ ('iand', (invert[left], a, b), (invert[right], c, d))))
+ optimizations.append((('inot', ('iand(is_used_once)', (left, a, b), (right, c, d))),
+ ('ior', (invert[left], a, b), (invert[right], c, d))))
+
+# Optimize x2bN(b2x(x)) -> x
+for size in type_sizes('bool'):
+ aN = 'a@' + str(size)
+ f2bN = 'f2b' + str(size)
+ i2bN = 'i2b' + str(size)
+ optimizations.append(((f2bN, ('b2f', aN)), a))
+ optimizations.append(((i2bN, ('b2i', aN)), a))
+
+# Optimize x2yN(b2x(x)) -> b2y
+for x, y in itertools.product(['f', 'u', 'i'], ['f', 'u', 'i']):
+ if x != 'f' and y != 'f' and x != y:
+ continue
+
+ b2x = 'b2f' if x == 'f' else 'b2i'
+ b2y = 'b2f' if y == 'f' else 'b2i'
+ x2yN = '{}2{}'.format(x, y)
+ optimizations.append(((x2yN, (b2x, a)), (b2y, a)))
+
+# Optimize away x2xN(a@N)
+for t in ['int', 'uint', 'float']:
+ for N in type_sizes(t):
+ x2xN = '{0}2{0}{1}'.format(t[0], N)
+ aN = 'a@{0}'.format(N)
+ optimizations.append(((x2xN, aN), a))
+
+# Optimize x2xN(y2yM(a@P)) -> y2yN(a) for integers
+# In particular, we can optimize away everything except upcast of downcast and
+# upcasts where the type differs from the other cast
+for N, M in itertools.product(type_sizes('uint'), type_sizes('uint')):
+ if N < M:
+ # The outer cast is a down-cast. It doesn't matter what the size of the
+ # argument of the inner cast is because we'll never been in the upcast
+ # of downcast case. Regardless of types, we'll always end up with y2yN
+ # in the end.
+ for x, y in itertools.product(['i', 'u'], ['i', 'u']):
+ x2xN = '{0}2{0}{1}'.format(x, N)
+ y2yM = '{0}2{0}{1}'.format(y, M)
+ y2yN = '{0}2{0}{1}'.format(y, N)
+ optimizations.append(((x2xN, (y2yM, a)), (y2yN, a)))
+ elif N > M:
+ # If the outer cast is an up-cast, we have to be more careful about the
+ # size of the argument of the inner cast and with types. In this case,
+ # the type is always the type of type up-cast which is given by the
+ # outer cast.
+ for P in type_sizes('uint'):
+ # We can't optimize away up-cast of down-cast.
+ if M < P:
+ continue
+
+ # Because we're doing down-cast of down-cast, the types always have
+ # to match between the two casts
+ for x in ['i', 'u']:
+ x2xN = '{0}2{0}{1}'.format(x, N)
+ x2xM = '{0}2{0}{1}'.format(x, M)
+ aP = 'a@{0}'.format(P)
+ optimizations.append(((x2xN, (x2xM, aP)), (x2xN, a)))
+ else:
+ # The N == M case is handled by other optimizations
+ pass
def fexp2i(exp, bits):
# We assume that exp is already in the right range.
- if bits == 32:
+ if bits == 16:
+ return ('i2i16', ('ishl', ('iadd', exp, 15), 10))
+ elif bits == 32:
return ('ishl', ('iadd', exp, 127), 23)
elif bits == 64:
return ('pack_64_2x32_split', 0, ('ishl', ('iadd', exp, 1023), 20))
# handles a range on exp of [-252, 254] which allows you to create any
# value (including denorms if the hardware supports it) and to adjust the
# exponent of any normal value to anything you want.
- if bits == 32:
+ if bits == 16:
+ exp = ('imin', ('imax', exp, -28), 30)
+ elif bits == 32:
exp = ('imin', ('imax', exp, -252), 254)
elif bits == 64:
exp = ('imin', ('imax', exp, -2044), 2046)
return ('fmul', ('fmul', f, pow2_1), pow2_2)
optimizations += [
+ (('ldexp@16', 'x', 'exp'), ldexp('x', 'exp', 16), 'options->lower_ldexp'),
(('ldexp@32', 'x', 'exp'), ldexp('x', 'exp', 32), 'options->lower_ldexp'),
(('ldexp@64', 'x', 'exp'), ldexp('x', 'exp', 64), 'options->lower_ldexp'),
]
((op, ('bcsel(is_used_once)', a, '#b', c), '#d'), ('bcsel', a, (op, b, d), (op, c, d)))
]
+# For derivatives in compute shaders, GLSL_NV_compute_shader_derivatives
+# states:
+#
+# If neither layout qualifier is specified, derivatives in compute shaders
+# return zero, which is consistent with the handling of built-in texture
+# functions like texture() in GLSL 4.50 compute shaders.
+for op in ['fddx', 'fddx_fine', 'fddx_coarse',
+ 'fddy', 'fddy_fine', 'fddy_coarse']:
+ optimizations += [
+ ((op, 'a'), 0.0, 'info->stage == MESA_SHADER_COMPUTE && info->cs.derivative_group == DERIVATIVE_GROUP_NONE')
+]
+
# This section contains "late" optimizations that should be run before
# creating ffmas and calling regular optimizations for the final time.
# Optimizations should go here if they help code generation and conflict
(('iadd', a, ('iadd', ('ineg', a), b)), b),
(('~fadd', ('fneg', a), ('fadd', a, b)), b),
(('~fadd', a, ('fadd', ('fneg', a), b)), b),
+
+ (('~flrp@32', ('fadd(is_used_once)', a, -1.0), ('fadd(is_used_once)', a, 1.0), d), ('fadd', ('flrp', -1.0, 1.0, d), a)),
+ (('~flrp@32', ('fadd(is_used_once)', a, 1.0), ('fadd(is_used_once)', a, -1.0), d), ('fadd', ('flrp', 1.0, -1.0, d), a)),
+ (('~flrp@32', ('fadd(is_used_once)', a, '#b'), ('fadd(is_used_once)', a, '#c'), d), ('fadd', ('fmul', d, ('fadd', c, ('fneg', b))), ('fadd', a, b))),
]
# This section contains "late" optimizations that should be run after the
(('fdot4', a, b), ('fdot_replicated4', a, b), 'options->fdot_replicates'),
(('fdph', a, b), ('fdph_replicated', a, b), 'options->fdot_replicates'),
- (('b2f(is_used_more_than_once)', ('inot', a)), ('bcsel', a, 0.0, 1.0)),
- (('fneg(is_used_more_than_once)', ('b2f', ('inot', a))), ('bcsel', a, -0.0, -1.0)),
+ (('~flrp@32', ('fadd(is_used_once)', a, b), ('fadd(is_used_once)', a, c), d), ('fadd', ('flrp', b, c, d), a)),
+ (('~flrp@64', ('fadd(is_used_once)', a, b), ('fadd(is_used_once)', a, c), d), ('fadd', ('flrp', b, c, d), a)),
+
+ (('b2f(is_used_more_than_once)', ('inot', 'a@1')), ('bcsel', a, 0.0, 1.0)),
+ (('fneg(is_used_more_than_once)', ('b2f', ('inot', 'a@1'))), ('bcsel', a, -0.0, -1.0)),
# we do these late so that we don't get in the way of creating ffmas
(('fmin', ('fadd(is_used_once)', '#c', a), ('fadd(is_used_once)', '#c', b)), ('fadd', c, ('fmin', a, b))),
(('fmax', ('fadd(is_used_once)', '#c', a), ('fadd(is_used_once)', '#c', b)), ('fadd', c, ('fmax', a, b))),
+
+ (('bcsel', a, 0, ('b2f32', ('inot', 'b@bool'))), ('b2f32', ('inot', ('ior', a, b)))),
]
-print nir_algebraic.AlgebraicPass("nir_opt_algebraic", optimizations).render()
-print nir_algebraic.AlgebraicPass("nir_opt_algebraic_before_ffma",
- before_ffma_optimizations).render()
-print nir_algebraic.AlgebraicPass("nir_opt_algebraic_late",
- late_optimizations).render()
+print(nir_algebraic.AlgebraicPass("nir_opt_algebraic", optimizations).render())
+print(nir_algebraic.AlgebraicPass("nir_opt_algebraic_before_ffma",
+ before_ffma_optimizations).render())
+print(nir_algebraic.AlgebraicPass("nir_opt_algebraic_late",
+ late_optimizations).render())
projects / mesa.git / blobdiff