optimizations = [
- (('imul', a, '#b@32(is_pos_power_of_two)'), ('ishl', a, ('find_lsb', b)), '!options->lower_bitops'),
- (('imul', a, '#b@32(is_neg_power_of_two)'), ('ineg', ('ishl', a, ('find_lsb', ('iabs', b)))), '!options->lower_bitops'),
- (('ishl', a, '#b@32'), ('imul', a, ('ishl', 1, b)), 'options->lower_bitops'),
+ (('imul', a, '#b(is_pos_power_of_two)'), ('ishl', a, ('find_lsb', b)), '!options->lower_bitops'),
+ (('imul', a, '#b(is_neg_power_of_two)'), ('ineg', ('ishl', a, ('find_lsb', ('iabs', b)))), '!options->lower_bitops'),
+ (('ishl', a, '#b'), ('imul', a, ('ishl', 1, b)), 'options->lower_bitops'),
(('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)),
(('idiv', a, 1), a),
(('umod', a, 1), 0),
(('imod', a, 1), 0),
- (('udiv', a, '#b@32(is_pos_power_of_two)'), ('ushr', a, ('find_lsb', b)), '!options->lower_bitops'),
- (('idiv', a, '#b@32(is_pos_power_of_two)'), ('imul', ('isign', a), ('ushr', ('iabs', a), ('find_lsb', b))), 'options->lower_idiv'),
- (('idiv', a, '#b@32(is_neg_power_of_two)'), ('ineg', ('imul', ('isign', a), ('ushr', ('iabs', a), ('find_lsb', ('iabs', b))))), 'options->lower_idiv'),
+ (('udiv', a, '#b(is_pos_power_of_two)'), ('ushr', a, ('find_lsb', b)), '!options->lower_bitops'),
+ (('idiv', a, '#b(is_pos_power_of_two)'), ('imul', ('isign', a), ('ushr', ('iabs', a), ('find_lsb', b))), 'options->lower_idiv'),
+ (('idiv', a, '#b(is_neg_power_of_two)'), ('ineg', ('imul', ('isign', a), ('ushr', ('iabs', a), ('find_lsb', ('iabs', b))))), 'options->lower_idiv'),
(('umod', a, '#b(is_pos_power_of_two)'), ('iand', a, ('isub', b, 1))),
(('~fneg', ('fneg', a)), a),
(('usadd_4x8', a, ~0), ~0),
(('~fadd', ('fmul', a, b), ('fmul', a, c)), ('fmul', a, ('fadd', b, c))),
(('iadd', ('imul', a, b), ('imul', a, c)), ('imul', a, ('iadd', b, c))),
+ (('iand', ('ior', a, b), ('ior', a, c)), ('ior', a, ('iand', b, c))),
+ (('ior', ('iand', a, b), ('iand', a, c)), ('iand', a, ('ior', b, c))),
(('~fadd', ('fneg', a), a), 0.0),
(('iadd', ('ineg', a), a), 0),
(('iadd', ('ineg', a), ('iadd', a, b)), b),
(('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'),
+ (('~fmul', ('fadd', ('iand', ('ineg', ('b2i', 'a@bool')), ('fmul', b, c)), '#d'), '#e'),
('bcsel', a, ('fmul', ('fadd', ('fmul', b, c), d), e), ('fmul', d, e))),
(('fdph', a, b), ('fdot4', ('vec4', 'a.x', 'a.y', 'a.z', 1.0), b), 'options->lower_fdph'),
optimizations.extend([
# 'a >> #b << #b' -> 'a & ~((1 << #b) - 1)'
- (('ishl@32', ('ushr@32', a, log2), log2), ('iand', a, mask)),
+ (('ishl', ('ushr', a, log2), log2), ('iand', a, mask)),
# Reassociate for improved CSE
- (('iand@32', ('iadd@32', a, b_is_multiple), mask), ('iadd', ('iand', a, mask), b)),
+ (('iand', ('iadd', a, b_is_multiple), mask), ('iadd', ('iand', a, mask), b)),
])
# To save space in the state tables, reduce to the set that is known to help.
# Comparison simplifications
(('~inot', ('flt', a, b)), ('fge', a, b)),
(('~inot', ('fge', a, b)), ('flt', a, b)),
- (('inot', ('feq', a, b)), ('fne', a, b)),
- (('inot', ('fne', a, b)), ('feq', a, b)),
+ (('inot', ('feq', a, b)), ('fneu', a, b)),
+ (('inot', ('fneu', 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', ('ieq', a, b)), ('ine', a, b)),
(('inot', ('ine', a, b)), ('ieq', a, b)),
- (('iand', ('feq', a, b), ('fne', a, b)), False),
+ (('iand', ('feq', a, b), ('fneu', a, b)), False),
(('iand', ('flt', a, b), ('flt', b, a)), False),
(('iand', ('ieq', a, b), ('ine', a, b)), False),
(('iand', ('ilt', a, b), ('ilt', b, a)), False),
(('flt', ('fneg', a), ('fneg', b)), ('flt', b, a)),
(('fge', ('fneg', a), ('fneg', b)), ('fge', b, a)),
(('feq', ('fneg', a), ('fneg', b)), ('feq', b, a)),
- (('fne', ('fneg', a), ('fneg', b)), ('fne', b, a)),
+ (('fneu', ('fneg', a), ('fneg', b)), ('fneu', b, a)),
(('flt', ('fneg', a), -1.0), ('flt', 1.0, a)),
(('flt', -1.0, ('fneg', a)), ('flt', a, 1.0)),
(('fge', ('fneg', a), -1.0), ('fge', 1.0, a)),
(('fge', -1.0, ('fneg', a)), ('fge', a, 1.0)),
- (('fne', ('fneg', a), -1.0), ('fne', 1.0, a)),
+ (('fneu', ('fneg', a), -1.0), ('fneu', 1.0, a)),
(('feq', -1.0, ('fneg', a)), ('feq', a, 1.0)),
- (('flt', ('fsat(is_used_once)', a), '#b(is_gt_0_and_lt_1)'), ('flt', a, b)),
+ # flt(fsat(a), b > 0 && b < 1) is inexact if a is NaN (fsat(NaN) is 0)
+ # because it returns True while flt(a, b) always returns False.
+ (('~flt', ('fsat(is_used_once)', a), '#b(is_gt_0_and_lt_1)'), ('flt', a, b)),
(('flt', '#b(is_gt_0_and_lt_1)', ('fsat(is_used_once)', a)), ('flt', b, a)),
(('fge', ('fsat(is_used_once)', a), '#b(is_gt_0_and_lt_1)'), ('fge', a, b)),
- (('fge', '#b(is_gt_0_and_lt_1)', ('fsat(is_used_once)', a)), ('fge', b, a)),
+ # fge(b > 0 && b < 1, fsat(a)) is inexact if a is NaN (fsat(NaN) is 0)
+ # because it returns True while fge(b, a) always returns False.
+ (('~fge', '#b(is_gt_0_and_lt_1)', ('fsat(is_used_once)', a)), ('fge', b, a)),
(('feq', ('fsat(is_used_once)', a), '#b(is_gt_0_and_lt_1)'), ('feq', a, b)),
- (('fne', ('fsat(is_used_once)', a), '#b(is_gt_0_and_lt_1)'), ('fne', a, b)),
+ (('fneu', ('fsat(is_used_once)', a), '#b(is_gt_0_and_lt_1)'), ('fneu', a, b)),
(('fge', ('fsat(is_used_once)', a), 1.0), ('fge', a, 1.0)),
(('flt', ('fsat(is_used_once)', a), 1.0), ('flt', a, 1.0)),
(('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)),
+ (('fneu', ('fadd', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('ior', a, b)),
+ (('fneu', ('fmax', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('ior', a, b)),
+ (('fneu', ('bcsel', a, 1.0, ('b2f', 'b@1')) , 0.0), ('ior', a, b)),
+ (('fneu', ('b2f', 'a@1'), ('fneg', ('b2f', 'b@1'))), ('ior', a, b)),
+ (('fneu', ('fmul', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('iand', a, b)),
+ (('fneu', ('fmin', ('b2f', 'a@1'), ('b2f', 'b@1')), 0.0), ('iand', a, b)),
+ (('fneu', ('bcsel', a, ('b2f', 'b@1'), 0.0) , 0.0), ('iand', a, b)),
+ (('fneu', ('fadd', ('b2f', 'a@1'), ('fneg', ('b2f', 'b@1'))), 0.0), ('ixor', a, b)),
+ (('fneu', ('b2f', 'a@1') , ('b2f', 'b@1') ), ('ixor', a, b)),
+ (('fneu', ('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))),
(('~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)),
+ (('~fneu', ('fadd', a, b), a), ('fneu', b, 0.0)),
(('~flt', ('fadd(is_used_once)', a, '#b'), '#c'), ('flt', a, ('fadd', c, ('fneg', b)))),
(('~flt', ('fneg(is_used_once)', ('fadd(is_used_once)', a, '#b')), '#c'), ('flt', ('fneg', ('fadd', c, b)), a)),
(('~fge', ('fadd(is_used_once)', a, '#b'), '#c'), ('fge', a, ('fadd', c, ('fneg', b)))),
(('~fge', ('fneg(is_used_once)', ('fadd(is_used_once)', a, '#b')), '#c'), ('fge', ('fneg', ('fadd', c, b)), a)),
(('~feq', ('fadd(is_used_once)', a, '#b'), '#c'), ('feq', a, ('fadd', c, ('fneg', b)))),
(('~feq', ('fneg(is_used_once)', ('fadd(is_used_once)', a, '#b')), '#c'), ('feq', ('fneg', ('fadd', c, b)), a)),
- (('~fne', ('fadd(is_used_once)', a, '#b'), '#c'), ('fne', a, ('fadd', c, ('fneg', b)))),
- (('~fne', ('fneg(is_used_once)', ('fadd(is_used_once)', a, '#b')), '#c'), ('fne', ('fneg', ('fadd', c, b)), a)),
+ (('~fneu', ('fadd(is_used_once)', a, '#b'), '#c'), ('fneu', a, ('fadd', c, ('fneg', b)))),
+ (('~fneu', ('fneg(is_used_once)', ('fadd(is_used_once)', a, '#b')), '#c'), ('fneu', ('fneg', ('fadd', c, b)), a)),
# Cannot remove the addition from ilt or ige due to overflow.
(('ieq', ('iadd', a, b), a), ('ieq', b, 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),
+ (('~fneu', ('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)),
+ (('fneu', ('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)),
+ (('fneu', ('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)),
# 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)), ('fneu', a, 0.0)),
# -fabs(a) < 0.0
# fabs(a) > 0.0
- (('~flt', ('fneg', ('fabs', a)), 0.0), ('fne', a, 0.0)),
+ (('~flt', ('fneg', ('fabs', a)), 0.0), ('fneu', a, 0.0)),
# 0.0 >= fabs(a)
# 0.0 == fabs(a) because fabs(a) must be >= 0
# !((a >= 0.0) && (a <= 1.0))
# !(a == fsat(a))
# a != fsat(a)
- (('ior', ('flt', a, 0.0), ('flt', 1.0, a)), ('fne', a, ('fsat', a)), '!options->lower_fsat'),
+ (('ior', ('flt', a, 0.0), ('flt', 1.0, a)), ('fneu', a, ('fsat', a)), '!options->lower_fsat'),
(('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)))),
(('iand@32', a, ('inot', ('ishr', a, 31))), ('imax', a, 0)),
# Simplify logic to detect sign of an integer.
- (('ieq', ('iand', a, 0x80000000), 0x00000000), ('ige', a, 0)),
- (('ine', ('iand', a, 0x80000000), 0x80000000), ('ige', a, 0)),
- (('ine', ('iand', a, 0x80000000), 0x00000000), ('ilt', a, 0)),
- (('ieq', ('iand', a, 0x80000000), 0x80000000), ('ilt', a, 0)),
+ (('ieq', ('iand', 'a@32', 0x80000000), 0x00000000), ('ige', a, 0)),
+ (('ine', ('iand', 'a@32', 0x80000000), 0x80000000), ('ige', a, 0)),
+ (('ine', ('iand', 'a@32', 0x80000000), 0x00000000), ('ilt', a, 0)),
+ (('ieq', ('iand', 'a@32', 0x80000000), 0x80000000), ('ilt', a, 0)),
(('ine', ('ushr', 'a@32', 31), 0), ('ilt', a, 0)),
(('ieq', ('ushr', 'a@32', 31), 0), ('ige', a, 0)),
(('ieq', ('ushr', 'a@32', 31), 1), ('ilt', a, 0)),
(('fmax', a, ('fabs', a)), ('fabs', a)),
(('imax', a, ('iabs', a)), ('iabs', a)),
(('fmax', a, ('fneg', a)), ('fabs', a)),
- (('imax', a, ('ineg', a)), ('iabs', a)),
+ (('imax', a, ('ineg', a)), ('iabs', a), '!options->lower_iabs'),
(('~fmax', ('fabs', a), 0.0), ('fabs', a)),
- (('~fmin', ('fmax', a, 0.0), 1.0), ('fsat', a), '!options->lower_fsat'),
+ (('fmin', ('fmax', a, 0.0), 1.0), ('fsat', a), '!options->lower_fsat'),
+ # fmax(fmin(a, 1.0), 0.0) is inexact because it returns 1.0 on NaN, while
+ # fsat(a) returns 0.0.
(('~fmax', ('fmin', a, 1.0), 0.0), ('fsat', a), '!options->lower_fsat'),
+ # fmin(fmax(a, -1.0), 0.0) is inexact because it returns -1.0 on NaN, while
+ # fneg(fsat(fneg(a))) returns -0.0 on NaN.
(('~fmin', ('fmax', a, -1.0), 0.0), ('fneg', ('fsat', ('fneg', a))), '!options->lower_fsat'),
+ # fmax(fmin(a, 0.0), -1.0) is inexact because it returns 0.0 on NaN, while
+ # fneg(fsat(fneg(a))) returns -0.0 on NaN. This only matters if
+ # SignedZeroInfNanPreserve is set, but we don't currently have any way of
+ # representing this in the optimizations other than the usual ~.
(('~fmax', ('fmin', a, 0.0), -1.0), ('fneg', ('fsat', ('fneg', a))), '!options->lower_fsat'),
(('fsat', ('fsign', a)), ('b2f', ('flt', 0.0, a))),
(('fsat', ('b2f', a)), ('b2f', a)),
(('fmin', ('fmax', ('fmin', ('fmax', a, b), c), b), c), ('fmin', ('fmax', a, b), c)),
(('imin', ('imax', ('imin', ('imax', a, b), c), b), c), ('imin', ('imax', a, b), c)),
(('umin', ('umax', ('umin', ('umax', a, b), c), b), c), ('umin', ('umax', a, b), c)),
- (('fmax', ('fsat', a), '#b@32(is_zero_to_one)'), ('fsat', ('fmax', a, b))),
- (('fmin', ('fsat', a), '#b@32(is_zero_to_one)'), ('fsat', ('fmin', a, b))),
+ # Both the left and right patterns are "b" when isnan(a), so this is exact.
+ (('fmax', ('fsat', a), '#b(is_zero_to_one)'), ('fsat', ('fmax', a, b))),
+ # The left pattern is 0.0 when isnan(a) (because fmin(fsat(NaN), b) ->
+ # fmin(0.0, b)) while the right one is "b", so this optimization is inexact.
+ (('~fmin', ('fsat', a), '#b(is_zero_to_one)'), ('fsat', ('fmin', a, b))),
# If a in [0,b] then b-a is also in [0,b]. Since b in [0,1], max(b-a, 0) =
# fsat(b-a).
# If a > b, then b-a < 0 and max(b-a, 0) = fsat(b-a) = 0
#
# This should be NaN safe since max(NaN, 0) = fsat(NaN) = 0.
- (('fmax', ('fadd(is_used_once)', ('fneg', 'a(is_not_negative)'), '#b@32(is_zero_to_one)'), 0.0),
+ (('fmax', ('fadd(is_used_once)', ('fneg', 'a(is_not_negative)'), '#b(is_zero_to_one)'), 0.0),
('fsat', ('fadd', ('fneg', a), b)), '!options->lower_fsat'),
(('extract_u8', ('imin', ('imax', a, 0), 0xff), 0), ('imin', ('imax', a, 0), 0xff)),
(('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))),
+ (('ine', ('ineg', ('b2i', 'a@1')), ('ineg', ('b2i', 'b@1'))), ('ine', a, b)),
+ (('b2i', ('ine', 'a@1', 'b@1')), ('b2i', ('ixor', a, b))),
- (('iand', ('ieq', 'a@32', 0), ('ieq', 'b@32', 0)), ('ieq', ('ior', a, b), 0), '!options->lower_bitops'),
- (('ior', ('ine', 'a@32', 0), ('ine', 'b@32', 0)), ('ine', ('ior', a, b), 0), '!options->lower_bitops'),
+ (('iand', ('ieq', 'a@32', 0), ('ieq', 'b@32', 0)), ('ieq', ('umax', a, b), 0)),
+ (('ior', ('ieq', 'a@32', 0), ('ieq', 'b@32', 0)), ('ieq', ('umin', a, b), 0)),
+ (('iand', ('ine', 'a@32', 0), ('ine', 'b@32', 0)), ('ine', ('umin', a, b), 0)),
+ (('ior', ('ine', 'a@32', 0), ('ine', 'b@32', 0)), ('ine', ('umax', a, b), 0)),
# This pattern occurs coutresy of __flt64_nonnan in the soft-fp64 code.
# The first part of the iand comes from the !__feq64_nonnan.
(('slt', a, b), ('b2f', ('flt', a, b)), 'options->lower_scmp'),
(('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'),
+ (('sne', a, b), ('b2f', ('fneu', a, b)), 'options->lower_scmp'),
(('seq', ('seq', a, b), 1.0), ('seq', a, b)),
(('seq', ('sne', a, b), 1.0), ('sne', a, b)),
(('seq', ('slt', a, b), 1.0), ('slt', a, b)),
(('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)),
+
+ (('ball_iequal2', a, b), ('iand', ('ieq', 'a.x', 'b.x'), ('ieq', 'a.y', 'b.y')), 'options->lower_vector_cmp'),
+ (('ball_iequal3', a, b), ('iand', ('iand', ('ieq', 'a.x', 'b.x'), ('ieq', 'a.y', 'b.y')), ('ieq', 'a.z', 'b.z')), 'options->lower_vector_cmp'),
+ (('ball_iequal4', a, b), ('iand', ('iand', ('ieq', 'a.x', 'b.x'), ('ieq', 'a.y', 'b.y')), ('iand', ('ieq', 'a.z', 'b.z'), ('ieq', 'a.w', 'b.w'))), 'options->lower_vector_cmp'),
+
+ (('bany_inequal2', a, b), ('ior', ('ine', 'a.x', 'b.x'), ('ine', 'a.y', 'b.y')), 'options->lower_vector_cmp'),
+ (('bany_inequal3', a, b), ('ior', ('ior', ('ine', 'a.x', 'b.x'), ('ine', 'a.y', 'b.y')), ('ine', 'a.z', 'b.z')), 'options->lower_vector_cmp'),
+ (('bany_inequal4', a, b), ('ior', ('ior', ('ine', 'a.x', 'b.x'), ('ine', 'a.y', 'b.y')), ('ior', ('ine', 'a.z', 'b.z'), ('ine', 'a.w', 'b.w'))), 'options->lower_vector_cmp'),
+
+ (('ball_fequal2', a, b), ('iand', ('feq', 'a.x', 'b.x'), ('feq', 'a.y', 'b.y')), 'options->lower_vector_cmp'),
+ (('ball_fequal3', a, b), ('iand', ('iand', ('feq', 'a.x', 'b.x'), ('feq', 'a.y', 'b.y')), ('feq', 'a.z', 'b.z')), 'options->lower_vector_cmp'),
+ (('ball_fequal4', a, b), ('iand', ('iand', ('feq', 'a.x', 'b.x'), ('feq', 'a.y', 'b.y')), ('iand', ('feq', 'a.z', 'b.z'), ('feq', 'a.w', 'b.w'))), 'options->lower_vector_cmp'),
+
+ (('bany_fnequal2', a, b), ('ior', ('fneu', 'a.x', 'b.x'), ('fneu', 'a.y', 'b.y')), 'options->lower_vector_cmp'),
+ (('bany_fnequal3', a, b), ('ior', ('ior', ('fneu', 'a.x', 'b.x'), ('fneu', 'a.y', 'b.y')), ('fneu', 'a.z', 'b.z')), 'options->lower_vector_cmp'),
+ (('bany_fnequal4', a, b), ('ior', ('ior', ('fneu', 'a.x', 'b.x'), ('fneu', 'a.y', 'b.y')), ('ior', ('fneu', 'a.z', 'b.z'), ('fneu', 'a.w', 'b.w'))), 'options->lower_vector_cmp'),
+
+ (('fneu', ('fneg', a), a), ('fneu', a, 0.0)),
(('feq', ('fneg', a), a), ('feq', a, 0.0)),
# Emulating booleans
(('imul', ('b2i', 'a@1'), ('b2i', 'b@1')), ('b2i', ('iand', a, b))),
+ (('iand', ('b2i', 'a@1'), ('b2i', 'b@1')), ('b2i', ('iand', a, b))),
+ (('ior', ('b2i', 'a@1'), ('b2i', 'b@1')), ('b2i', ('ior', 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)),
(('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), ('b2f', ('inot', a))),
- (('bcsel@32', a, -1.0, -0.0), ('fneg', ('b2f', a))),
- (('bcsel@32', a, -0.0, -1.0), ('fneg', ('b2f', ('inot', a)))),
+ (('bcsel', a, 1.0, 0.0), ('b2f', a)),
+ (('bcsel', a, 0.0, 1.0), ('b2f', ('inot', a))),
+ (('bcsel', a, -1.0, -0.0), ('fneg', ('b2f', a))),
+ (('bcsel', 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))),
# D3D Boolean emulation
(('bcsel', a, -1, 0), ('ineg', ('b2i', 'a@1'))),
(('bcsel', a, 0, -1), ('ineg', ('b2i', ('inot', a)))),
+ (('bcsel', a, 1, 0), ('b2i', 'a@1')),
+ (('bcsel', a, 0, 1), ('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'))),
(('~f2u32', ('i2f', 'a@32')), a),
(('~f2u32', ('u2f', 'a@32')), a),
- # Conversions from float16 to float32 and back can always be removed
+ # Conversions from 16 bits to 32 bits and back can always be removed
(('f2f16', ('f2f32', 'a@16')), a),
(('f2fmp', ('f2f32', 'a@16')), a),
- # Conversions to float16 would be lossy so they should only be removed if
+ (('i2i16', ('i2i32', 'a@16')), a),
+ (('i2imp', ('i2i32', 'a@16')), a),
+ (('u2u16', ('u2u32', 'a@16')), a),
+ (('u2ump', ('u2u32', 'a@16')), a),
+ (('f2f16', ('b2f32', 'a@1')), ('b2f16', a)),
+ (('f2fmp', ('b2f32', 'a@1')), ('b2f16', a)),
+ (('i2i16', ('b2i32', 'a@1')), ('b2i16', a)),
+ (('i2imp', ('b2i32', 'a@1')), ('b2i16', a)),
+ (('u2u16', ('b2i32', 'a@1')), ('b2i16', a)),
+ (('u2ump', ('b2i32', 'a@1')), ('b2i16', a)),
+ # Conversions to 16 bits would be lossy so they should only be removed if
# the instruction was generated by the precision lowering pass.
(('f2f32', ('f2fmp', 'a@32')), a),
+ (('i2i32', ('i2imp', 'a@32')), a),
+ (('u2u32', ('u2ump', 'a@32')), a),
(('ffloor', 'a(is_integral)'), a),
(('fceil', 'a(is_integral)'), a),
(('fmax', ('fadd', ('fmul', ('fsat', a), ('fneg', ('fsat', a))), 1.0), 0.0), ('fadd', ('fmul', ('fsat', a), ('fneg', ('fsat', a))), 1.0)),
(('fmax', ('fadd', ('fneg', ('fmul', ('fsat', a), ('fsat', a))), 1.0), 0.0), ('fadd', ('fneg', ('fmul', ('fsat', a), ('fsat', a))), 1.0)),
- (('fne', 'a(is_not_zero)', 0.0), True),
+ (('fneu', 'a(is_not_zero)', 0.0), True),
(('feq', 'a(is_not_zero)', 0.0), False),
# In this chart, + means value > 0 and - means value < 0.
# Packing and then unpacking does nothing
(('unpack_64_2x32_split_x', ('pack_64_2x32_split', a, b)), a),
(('unpack_64_2x32_split_y', ('pack_64_2x32_split', a, b)), b),
+ (('unpack_64_2x32', ('pack_64_2x32_split', a, b)), ('vec2', a, b)),
+ (('unpack_64_2x32', ('pack_64_2x32', a)), a),
(('pack_64_2x32_split', ('unpack_64_2x32_split_x', a),
('unpack_64_2x32_split_y', a)), a),
+ (('pack_64_2x32', ('vec2', ('unpack_64_2x32_split_x', a),
+ ('unpack_64_2x32_split_y', a))), a),
+ (('pack_64_2x32', ('unpack_64_2x32', a)), a),
# Comparing two halves of an unpack separately. While this optimization
# should be correct for non-constant values, it's less obvious that it's
(('ishr', 'a@64', 56), ('extract_i8', a, 7), '!options->lower_extract_byte'),
(('iand', 0xff, a), ('extract_u8', a, 0), '!options->lower_extract_byte'),
+ (('ubfe', a, 0, 8), ('extract_u8', a, 0), '!options->lower_extract_byte'),
+ (('ubfe', a, 8, 8), ('extract_u8', a, 1), '!options->lower_extract_byte'),
+ (('ubfe', a, 16, 8), ('extract_u8', a, 2), '!options->lower_extract_byte'),
+ (('ubfe', a, 24, 8), ('extract_u8', a, 3), '!options->lower_extract_byte'),
+ (('ibfe', a, 0, 8), ('extract_i8', a, 0), '!options->lower_extract_byte'),
+ (('ibfe', a, 8, 8), ('extract_i8', a, 1), '!options->lower_extract_byte'),
+ (('ibfe', a, 16, 8), ('extract_i8', a, 2), '!options->lower_extract_byte'),
+ (('ibfe', a, 24, 8), ('extract_i8', a, 3), '!options->lower_extract_byte'),
+
+ # Word extraction
+ (('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'),
+
+ (('ubfe', a, 0, 16), ('extract_u16', a, 0), '!options->lower_extract_word'),
+ (('ubfe', a, 16, 16), ('extract_u16', a, 1), '!options->lower_extract_word'),
+ (('ibfe', a, 0, 16), ('extract_i16', a, 0), '!options->lower_extract_word'),
+ (('ibfe', a, 16, 16), ('extract_i16', a, 1), '!options->lower_extract_word'),
+
+ # Lower pack/unpack
+ (('pack_64_2x32_split', a, b), ('ior', ('u2u64', a), ('ishl', ('u2u64', b), 32)), 'options->lower_pack_64_2x32_split'),
+ (('pack_32_2x16_split', a, b), ('ior', ('u2u32', a), ('ishl', ('u2u32', b), 16)), 'options->lower_pack_32_2x16_split'),
+ (('unpack_64_2x32_split_x', a), ('u2u32', a), 'options->lower_unpack_64_2x32_split'),
+ (('unpack_64_2x32_split_y', a), ('u2u32', ('ushr', a, 32)), 'options->lower_unpack_64_2x32_split'),
+ (('unpack_32_2x16_split_x', a), ('u2u16', a), 'options->lower_unpack_32_2x16_split'),
+ (('unpack_32_2x16_split_y', a), ('u2u16', ('ushr', a, 16)), 'options->lower_unpack_32_2x16_split'),
+
# Useless masking before unpacking
(('unpack_half_2x16_split_x', ('iand', a, 0xffff)), ('unpack_half_2x16_split_x', a)),
(('unpack_32_2x16_split_x', ('iand', a, 0xffff)), ('unpack_32_2x16_split_x', a)),
(('unpack_32_2x16_split_y', ('iand', a, 0xffff0000)), ('unpack_32_2x16_split_y', a)),
(('unpack_64_2x32_split_y', ('iand', a, 0xffffffff00000000)), ('unpack_64_2x32_split_y', a)),
+ (('unpack_half_2x16_split_x', ('extract_u16', a, 0)), ('unpack_half_2x16_split_x', a)),
+ (('unpack_half_2x16_split_x', ('extract_u16', a, 1)), ('unpack_half_2x16_split_y', a)),
+ (('unpack_32_2x16_split_x', ('extract_u16', a, 0)), ('unpack_32_2x16_split_x', a)),
+ (('unpack_32_2x16_split_x', ('extract_u16', a, 1)), ('unpack_32_2x16_split_y', a)),
+
# Optimize half packing
(('ishl', ('pack_half_2x16', ('vec2', a, 0)), 16), ('pack_half_2x16', ('vec2', 0, a))),
- (('ishr', ('pack_half_2x16', ('vec2', 0, a)), 16), ('pack_half_2x16', ('vec2', a, 0))),
+ (('ushr', ('pack_half_2x16', ('vec2', 0, a)), 16), ('pack_half_2x16', ('vec2', a, 0))),
(('iadd', ('pack_half_2x16', ('vec2', a, 0)), ('pack_half_2x16', ('vec2', 0, b))),
('pack_half_2x16', ('vec2', a, b))),
(('ior', ('pack_half_2x16', ('vec2', a, 0)), ('pack_half_2x16', ('vec2', 0, b))),
('pack_half_2x16', ('vec2', a, b))),
+
+ (('ishl', ('pack_half_2x16_split', a, 0), 16), ('pack_half_2x16_split', 0, a)),
+ (('ushr', ('pack_half_2x16_split', 0, a), 16), ('pack_half_2x16_split', a, 0)),
+ (('extract_u16', ('pack_half_2x16_split', 0, a), 1), ('pack_half_2x16_split', a, 0)),
+
+ (('iadd', ('pack_half_2x16_split', a, 0), ('pack_half_2x16_split', 0, b)), ('pack_half_2x16_split', a, b)),
+ (('ior', ('pack_half_2x16_split', a, 0), ('pack_half_2x16_split', 0, b)), ('pack_half_2x16_split', a, b)),
])
# After the ('extract_u8', a, 0) pattern, above, triggers, there will be
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', ('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
(('ussub_4x8', a, 0), a),
(('ussub_4x8', a, ~0), 0),
(('bcsel', ('ine', a, -1), ('ifind_msb', a), -1), ('ifind_msb', a)),
- (('fmin3@64', a, b, c), ('fmin@64', a, ('fmin@64', b, c))),
- (('fmax3@64', a, b, c), ('fmax@64', a, ('fmax@64', b, c))),
- (('fmed3@64', a, b, c), ('fmax@64', ('fmin@64', ('fmax@64', a, b), c), ('fmin@64', a, b))),
+ (('~fmul', ('bcsel(is_used_once)', c, -1.0, 1.0), b), ('bcsel', c, ('fneg', b), b)),
+ (('~fmul', ('bcsel(is_used_once)', c, 1.0, -1.0), b), ('bcsel', c, b, ('fneg', b))),
+ (('~bcsel', ('flt', a, 0.0), ('fneg', a), a), ('fabs', a)),
+
+ (('bcsel', a, ('bcsel', b, c, d), d), ('bcsel', ('iand', a, b), c, d)),
+ (('bcsel', a, b, ('bcsel', c, b, d)), ('bcsel', ('ior', a, c), b, d)),
# Misc. lowering
(('fmod', a, b), ('fsub', a, ('fmul', b, ('ffloor', ('fdiv', a, b)))), 'options->lower_fmod'),
(('frem', a, b), ('fsub', a, ('fmul', b, ('ftrunc', ('fdiv', a, b)))), 'options->lower_fmod'),
- (('uadd_carry@32', a, b), ('b2i', ('ult', ('iadd', a, b), a)), 'options->lower_uadd_carry'),
+ (('uadd_carry', a, b), ('b2i', ('ult', ('iadd', a, b), a)), 'options->lower_uadd_carry'),
(('usub_borrow@32', a, b), ('b2i', ('ult', a, b)), 'options->lower_usub_borrow'),
(('bitfield_insert', 'base', 'insert', 'offset', 'bits'),
(('bfm', 'bits', ('iand', 31, 'offset')), ('bfm', 'bits', 'offset')),
(('bfm', ('iand', 31, 'bits'), 'offset'), ('bfm', 'bits', 'offset')),
+ # Section 8.8 (Integer Functions) of the GLSL 4.60 spec says:
+ #
+ # If bits is zero, the result will be zero.
+ #
+ # These patterns prevent other patterns from generating invalid results
+ # when count is zero.
+ (('ubfe', a, b, 0), 0),
+ (('ibfe', a, b, 0), 0),
+
+ (('ubfe', a, 0, '#b'), ('iand', a, ('ushr', 0xffffffff, ('ineg', b)))),
+
+ (('b2i32', ('i2b', ('ubfe', a, b, 1))), ('ubfe', a, b, 1)),
+ (('b2i32', ('i2b', ('ibfe', a, b, 1))), ('ubfe', a, b, 1)), # ubfe in the replacement is correct
+ (('ine', ('ibfe(is_used_once)', a, '#b', '#c'), 0), ('ine', ('iand', a, ('ishl', ('ushr', 0xffffffff, ('ineg', c)), b)), 0)),
+ (('ieq', ('ibfe(is_used_once)', a, '#b', '#c'), 0), ('ieq', ('iand', a, ('ishl', ('ushr', 0xffffffff, ('ineg', c)), b)), 0)),
+ (('ine', ('ubfe(is_used_once)', a, '#b', '#c'), 0), ('ine', ('iand', a, ('ishl', ('ushr', 0xffffffff, ('ineg', c)), b)), 0)),
+ (('ieq', ('ubfe(is_used_once)', a, '#b', '#c'), 0), ('ieq', ('iand', a, ('ishl', ('ushr', 0xffffffff, ('ineg', c)), b)), 0)),
+
(('ibitfield_extract', 'value', 'offset', 'bits'),
('bcsel', ('ieq', 0, 'bits'),
0,
(('pack_half_2x16_split', 'a@32', 'b@32'),
('ior', ('ishl', ('u2u32', ('f2f16', b)), 16), ('u2u32', ('f2f16', a))),
- 'options->lower_pack_half_2x16_split'),
+ 'options->lower_pack_split'),
(('unpack_half_2x16_split_x', 'a@32'),
('f2f32', ('u2u16', a)),
- 'options->lower_unpack_half_2x16_split'),
+ 'options->lower_pack_split'),
(('unpack_half_2x16_split_y', 'a@32'),
('f2f32', ('u2u16', ('ushr', a, 16))),
- 'options->lower_unpack_half_2x16_split'),
+ 'options->lower_pack_split'),
+
+ (('pack_32_2x16_split', 'a@16', 'b@16'),
+ ('ior', ('ishl', ('u2u32', b), 16), ('u2u32', a)),
+ 'options->lower_pack_split'),
+
+ (('unpack_32_2x16_split_x', 'a@32'),
+ ('u2u16', a),
+ 'options->lower_pack_split'),
+
+ (('unpack_32_2x16_split_y', 'a@32'),
+ ('u2u16', ('ushr', 'a', 16)),
+ 'options->lower_pack_split'),
(('isign', a), ('imin', ('imax', a, -1), 1), 'options->lower_isign'),
+ (('imin', ('imax', a, -1), 1), ('isign', a), '!options->lower_isign'),
+ (('imax', ('imin', a, 1), -1), ('isign', a), '!options->lower_isign'),
(('fsign', a), ('fsub', ('b2f', ('flt', 0.0, a)), ('b2f', ('flt', a, 0.0))), 'options->lower_fsign'),
+ (('fadd', ('b2f32', ('flt', 0.0, 'a@32')), ('fneg', ('b2f32', ('flt', 'a@32', 0.0)))), ('fsign', a), '!options->lower_fsign'),
+ (('iadd', ('b2i32', ('flt', 0, 'a@32')), ('ineg', ('b2i32', ('flt', 'a@32', 0)))), ('f2i32', ('fsign', a)), '!options->lower_fsign'),
# Address/offset calculations:
# Drivers supporting imul24 should use the nir_lower_amul() pass, this
# rule converts everyone else to imul:
(('amul', a, b), ('imul', a, b), '!options->has_imul24'),
+ (('umul24', a, b),
+ ('imul', ('iand', a, 0xffffff), ('iand', b, 0xffffff)),
+ '!options->has_umul24'),
+ (('umad24', a, b, c),
+ ('iadd', ('imul', ('iand', a, 0xffffff), ('iand', b, 0xffffff)), c),
+ '!options->has_umad24'),
+
(('imad24_ir3', a, b, 0), ('imul24', a, b)),
(('imad24_ir3', a, 0, c), (c)),
(('imad24_ir3', a, 1, c), ('iadd', a, c)),
('bcsel', ('ilt', a, ('isub', a, b)), intmin, ('isub', a, b))), 'options->lower_add_sat'),
]
-invert = OrderedDict([('feq', 'fne'), ('fne', 'feq')])
+invert = OrderedDict([('feq', 'fneu'), ('fneu', 'feq')])
for left, right in itertools.combinations_with_replacement(invert.keys(), 2):
optimizations.append((('inot', ('ior(is_used_once)', (left, a, b), (right, c, d))),
if N == 1 or N >= M:
continue
+ cond = 'true'
+ if N == 8:
+ cond = 'options->support_8bit_alu'
+ elif N == 16:
+ cond = 'options->support_16bit_alu'
x2xM = '{0}2{0}{1}'.format(t[0], M)
x2xN = '{0}2{0}{1}'.format(t[0], N)
aN = 'a@' + str(N)
bN = 'b@' + str(N)
xeq = 'feq' if t == 'float' else 'ieq'
- xne = 'fne' if t == 'float' else 'ine'
+ xne = 'fneu' if t == 'float' else 'ine'
xge = '{0}ge'.format(t[0])
xlt = '{0}lt'.format(t[0])
bP = 'b@' + str(P)
optimizations += [
- ((xeq, (x2xM, aN), (x2xM, bP)), (xeq, a, (x2xN, b))),
- ((xne, (x2xM, aN), (x2xM, bP)), (xne, a, (x2xN, b))),
- ((xge, (x2xM, aN), (x2xM, bP)), (xge, a, (x2xN, b))),
- ((xlt, (x2xM, aN), (x2xM, bP)), (xlt, a, (x2xN, b))),
- ((xge, (x2xM, bP), (x2xM, aN)), (xge, (x2xN, b), a)),
- ((xlt, (x2xM, bP), (x2xM, aN)), (xlt, (x2xN, b), a)),
+ ((xeq, (x2xM, aN), (x2xM, bP)), (xeq, a, (x2xN, b)), cond),
+ ((xne, (x2xM, aN), (x2xM, bP)), (xne, a, (x2xN, b)), cond),
+ ((xge, (x2xM, aN), (x2xM, bP)), (xge, a, (x2xN, b)), cond),
+ ((xlt, (x2xM, aN), (x2xM, bP)), (xlt, a, (x2xN, b)), cond),
+ ((xge, (x2xM, bP), (x2xM, aN)), (xge, (x2xN, b), a), cond),
+ ((xlt, (x2xM, bP), (x2xM, aN)), (xlt, (x2xN, b), a), cond),
]
# The next bit doesn't work on floats because the range checks would
# and a check that the constant fits in the smaller bit size.
optimizations += [
((xeq, (x2xM, aN), '#b'),
- ('iand', (xeq, a, (x2xN, b)), (xeq, (x2xM, (x2xN, b)), b))),
+ ('iand', (xeq, a, (x2xN, b)), (xeq, (x2xM, (x2xN, b)), b)), cond),
((xne, (x2xM, aN), '#b'),
- ('ior', (xne, a, (x2xN, b)), (xne, (x2xM, (x2xN, b)), b))),
+ ('ior', (xne, a, (x2xN, b)), (xne, (x2xM, (x2xN, b)), b)), cond),
((xlt, (x2xM, aN), '#b'),
('iand', (xlt, xN_min, b),
- ('ior', (xlt, xN_max, b), (xlt, a, (x2xN, b))))),
+ ('ior', (xlt, xN_max, b), (xlt, a, (x2xN, b)))), cond),
((xlt, '#a', (x2xM, bN)),
('iand', (xlt, a, xN_max),
- ('ior', (xlt, a, xN_min), (xlt, (x2xN, a), b)))),
+ ('ior', (xlt, a, xN_min), (xlt, (x2xN, a), b))), cond),
((xge, (x2xM, aN), '#b'),
('iand', (xge, xN_max, b),
- ('ior', (xge, xN_min, b), (xge, a, (x2xN, b))))),
+ ('ior', (xge, xN_min, b), (xge, a, (x2xN, b)))), cond),
((xge, '#a', (x2xM, bN)),
('iand', (xge, a, xN_min),
- ('ior', (xge, a, xN_max), (xge, (x2xN, a), b)))),
+ ('ior', (xge, a, xN_max), (xge, (x2xN, a), b))), cond),
]
+# Convert masking followed by signed downcast to just unsigned downcast
+optimizations += [
+ (('i2i32', ('iand', 'a@64', 0xffffffff)), ('u2u32', a)),
+ (('i2i16', ('iand', 'a@32', 0xffff)), ('u2u16', a)),
+ (('i2i16', ('iand', 'a@64', 0xffff)), ('u2u16', a)),
+ (('i2i8', ('iand', 'a@16', 0xff)), ('u2u8', a)),
+ (('i2i8', ('iand', 'a@32', 0xff)), ('u2u8', a)),
+ (('i2i8', ('iand', 'a@64', 0xff)), ('u2u8', a)),
+]
+
def fexp2i(exp, bits):
# Generate an expression which constructs value 2.0^exp or 0.0.
#
# which constant folding will eat for lunch. The resulting ternary will
# further get cleaned up by the boolean reductions above and we will be
# left with just the original variable "a".
-for op in ['flt', 'fge', 'feq', 'fne',
+for op in ['flt', 'fge', 'feq', 'fneu',
'ilt', 'ige', 'ieq', 'ine', 'ult', 'uge']:
optimizations += [
((op, ('bcsel', 'a', '#b', '#c'), '#d'),
(('bcsel', a, (op, b, c), (op + '(is_used_once)', d, c)), (op, ('bcsel', a, b, d), c)),
]
-for op in ['frcp', 'frsq', 'fsqrt', 'fexp2', 'flog2', 'fsign', 'fsin', 'fcos']:
+for op in ['frcp', 'frsq', 'fsqrt', 'fexp2', 'flog2', 'fsign', 'fsin', 'fcos', 'fneg', 'fabs', 'fsign']:
+ optimizations += [
+ (('bcsel', c, (op + '(is_used_once)', a), (op + '(is_used_once)', b)), (op, ('bcsel', c, a, b))),
+ ]
+
+for op in ['ineg', 'iabs', 'inot', 'isign']:
optimizations += [
- (('bcsel', a, (op + '(is_used_once)', b), (op, c)), (op, ('bcsel', a, b, c))),
- (('bcsel', a, (op, b), (op + '(is_used_once)', c)), (op, ('bcsel', a, b, c))),
+ ((op, ('bcsel', c, '#a', '#b')), ('bcsel', c, (op, a), (op, b))),
]
+# This section contains optimizations to propagate downsizing conversions of
+# constructed vectors into vectors of downsized components. Whether this is
+# useful depends on the SIMD semantics of the backend. On a true SIMD machine,
+# this reduces the register pressure of the vector itself and often enables the
+# conversions to be eliminated via other algebraic rules or constant folding.
+# In the worst case on a SIMD architecture, the propagated conversions may be
+# revectorized via nir_opt_vectorize so instruction count is minimally
+# impacted.
+#
+# On a machine with SIMD-within-a-register only, this actually
+# counterintuitively hurts instruction count. These machines are the same that
+# require vectorize_vec2_16bit, so we predicate the optimizations on that flag
+# not being set.
+#
+# Finally for scalar architectures, there should be no difference in generated
+# code since it all ends up scalarized at the end, but it might minimally help
+# compile-times.
+
+for i in range(2, 4 + 1):
+ for T in ('f', 'u', 'i'):
+ vec_inst = ('vec' + str(i),)
+
+ indices = ['a', 'b', 'c', 'd']
+ suffix_in = tuple((indices[j] + '@32') for j in range(i))
+
+ to_16 = '{}2{}16'.format(T, T)
+ to_mp = '{}2{}mp'.format(T, T)
+
+ out_16 = tuple((to_16, indices[j]) for j in range(i))
+ out_mp = tuple((to_mp, indices[j]) for j in range(i))
+
+ optimizations += [
+ ((to_16, vec_inst + suffix_in), vec_inst + out_16, '!options->vectorize_vec2_16bit'),
+ ((to_mp, vec_inst + suffix_in), vec_inst + out_mp, '!options->vectorize_vec2_16bit')
+ ]
+
# 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
(('~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))),
+ (('~flrp', ('fadd(is_used_once)', a, -1.0), ('fadd(is_used_once)', a, 1.0), d), ('fadd', ('flrp', -1.0, 1.0, d), a)),
+ (('~flrp', ('fadd(is_used_once)', a, 1.0), ('fadd(is_used_once)', a, -1.0), d), ('fadd', ('flrp', 1.0, -1.0, d), a)),
+ (('~flrp', ('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
(('~fge', ('fadd', a, b), 0.0), ('fge', a, ('fneg', b))),
(('~fge', ('fneg', ('fadd', a, b)), 0.0), ('fge', ('fneg', a), b)),
(('~feq', ('fadd', a, b), 0.0), ('feq', a, ('fneg', b))),
- (('~fne', ('fadd', a, b), 0.0), ('fne', a, ('fneg', b))),
+ (('~fneu', ('fadd', a, b), 0.0), ('fneu', a, ('fneg', b))),
# nir_lower_to_source_mods will collapse this, but its existence during the
# optimization loop can prevent other optimizations.
(('iadd', 'a', ('ineg', 'b')), ('isub', 'a', 'b'), '!options->lower_sub'),
(('fneg', a), ('fsub', 0.0, a), 'options->lower_negate'),
(('ineg', a), ('isub', 0, a), 'options->lower_negate'),
+ (('iabs', a), ('imax', a, ('ineg', a)), 'options->lower_iabs'),
# These are duplicated from the main optimizations table. The late
# patterns that rearrange expressions like x - .5 < 0 to x < .5 can create
(('fge', ('fsat(is_used_once)', a), '#b(is_gt_0_and_lt_1)'), ('fge', a, b)),
(('fge', '#b(is_gt_0_and_lt_1)', ('fsat(is_used_once)', a)), ('fge', b, a)),
(('feq', ('fsat(is_used_once)', a), '#b(is_gt_0_and_lt_1)'), ('feq', a, b)),
- (('fne', ('fsat(is_used_once)', a), '#b(is_gt_0_and_lt_1)'), ('fne', a, b)),
+ (('fneu', ('fsat(is_used_once)', a), '#b(is_gt_0_and_lt_1)'), ('fneu', a, b)),
(('fge', ('fsat(is_used_once)', a), 1.0), ('fge', a, 1.0)),
- (('flt', ('fsat(is_used_once)', a), 1.0), ('flt', a, 1.0)),
+ # flt(fsat(a), 1.0) is inexact because it returns True if a is NaN
+ # (fsat(NaN) is 0), while flt(a, 1.0) always returns FALSE.
+ (('~flt', ('fsat(is_used_once)', a), 1.0), ('flt', a, 1.0)),
(('~fge', ('fmin(is_used_once)', ('fadd(is_used_once)', a, b), ('fadd', c, d)), 0.0), ('iand', ('fge', a, ('fneg', b)), ('fge', c, ('fneg', d)))),
(('flt', ('fneg', a), ('fneg', b)), ('flt', b, a)),
(('fge', ('fneg', a), ('fneg', b)), ('fge', b, a)),
(('feq', ('fneg', a), ('fneg', b)), ('feq', b, a)),
- (('fne', ('fneg', a), ('fneg', b)), ('fne', b, a)),
+ (('fneu', ('fneg', a), ('fneg', b)), ('fneu', b, a)),
(('flt', ('fneg', a), -1.0), ('flt', 1.0, a)),
(('flt', -1.0, ('fneg', a)), ('flt', a, 1.0)),
(('fge', ('fneg', a), -1.0), ('fge', 1.0, a)),
(('fge', -1.0, ('fneg', a)), ('fge', a, 1.0)),
- (('fne', ('fneg', a), -1.0), ('fne', 1.0, a)),
+ (('fneu', ('fneg', a), -1.0), ('fneu', 1.0, a)),
(('feq', -1.0, ('fneg', a)), ('feq', a, 1.0)),
(('ior', a, a), a),
(('fdot4', a, b), ('fdot_replicated4', a, b), 'options->fdot_replicates'),
(('fdph', a, b), ('fdph_replicated', a, b), 'options->fdot_replicates'),
- (('~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)),
+ (('~flrp', ('fadd(is_used_once)', a, b), ('fadd(is_used_once)', a, c), d), ('fadd', ('flrp', b, c, d), a)),
(('~fadd@32', 1.0, ('fmul(is_used_once)', c , ('fadd', b, -1.0 ))), ('fadd', ('fadd', 1.0, ('fneg', c)), ('fmul', b, c)), 'options->lower_flrp32'),
(('~fadd@64', 1.0, ('fmul(is_used_once)', c , ('fadd', b, -1.0 ))), ('fadd', ('fadd', 1.0, ('fneg', c)), ('fmul', b, c)), 'options->lower_flrp64'),
(('~fadd', ('ffma(is_used_once)', a, b, ('fmul', 'c(is_not_const_and_not_fsign)', 'd(is_not_const_and_not_fsign)') ), 'e(is_not_const)'),
('ffma', a, b, ('ffma', c, d, e)), '(info->stage != MESA_SHADER_VERTEX && info->stage != MESA_SHADER_GEOMETRY) && !options->intel_vec4'),
- # Convert f2fmp instructions to concrete f2f16 instructions. At this point
+ # Convert *2*mp instructions to concrete *2*16 instructions. At this point
# any conversions that could have been removed will have been removed in
# nir_opt_algebraic so any remaining ones are required.
(('f2fmp', a), ('f2f16', a)),
+ (('i2imp', a), ('i2i16', a)),
+ (('u2ump', a), ('u2u16', a)),
+
+ # Section 8.8 (Integer Functions) of the GLSL 4.60 spec says:
+ #
+ # If bits is zero, the result will be zero.
+ #
+ # These prevent the next two lowerings generating incorrect results when
+ # count is zero.
+ (('ubfe', a, b, 0), 0),
+ (('ibfe', a, b, 0), 0),
+
+ # On Intel GPUs, BFE is a 3-source instruction. Like all 3-source
+ # instructions on Intel GPUs, it cannot have an immediate values as
+ # sources. There are also limitations on source register strides. As a
+ # result, it is very easy for 3-source instruction combined with either
+ # loads of immediate values or copies from weird register strides to be
+ # more expensive than the primitive instructions it represents.
+ (('ubfe', a, '#b', '#c'), ('iand', ('ushr', 0xffffffff, ('ineg', c)), ('ushr', a, b)), 'options->lower_bfe_with_two_constants'),
+
+ # b is the lowest order bit to be extracted and c is the number of bits to
+ # extract. The inner shift removes the bits above b + c by shifting left
+ # 32 - (b + c). ishl only sees the low 5 bits of the shift count, which is
+ # -(b + c). The outer shift moves the bit that was at b to bit zero.
+ # After the first shift, that bit is now at b + (32 - (b + c)) or 32 - c.
+ # This means that it must be shifted right by 32 - c or -c bits.
+ (('ibfe', a, '#b', '#c'), ('ishr', ('ishl', a, ('ineg', ('iadd', b, c))), ('ineg', c)), 'options->lower_bfe_with_two_constants'),
+
+ # Clean up no-op shifts that may result from the bfe lowerings.
+ (('ishl', a, 0), a),
+ (('ishl', a, -32), a),
+ (('ishr', a, 0), a),
+ (('ishr', a, -32), a),
+ (('ushr', a, 0), a),
]
for op in ['fadd']:
(('fdot_replicated4', ('fneg', a), ('fneg', b)), ('fdot_replicated4', a, b)),
(('fneg', ('fneg', a)), a),
+ (('fneg', ('fmul(is_used_once)', a, b)), ('fmul', ('fneg', a), b)),
+ (('fabs', ('fmul(is_used_once)', a, b)), ('fmul', ('fabs', a), ('fabs', b))),
+
(('fneg', ('ffma(is_used_once)', a, b, c)), ('ffma', ('fneg', a), b, ('fneg', c))),
(('fneg', ('flrp(is_used_once)', a, b, c)), ('flrp', ('fneg', a), ('fneg', b), c)),
(('fneg', ('fadd(is_used_once)', a, b)), ('fadd', ('fneg', a), ('fneg', b))),
(('fneg', ('fmin(is_used_once)', a, b)), ('fmax', ('fneg', a), ('fneg', b))),
(('fneg', ('fmax(is_used_once)', a, b)), ('fmin', ('fneg', a), ('fneg', b))),
+ (('fneg', ('fdot_replicated2(is_used_once)', a, b)), ('fdot_replicated2', ('fneg', a), b)),
+ (('fneg', ('fdot_replicated3(is_used_once)', a, b)), ('fdot_replicated3', ('fneg', a), b)),
+ (('fneg', ('fdot_replicated4(is_used_once)', a, b)), ('fdot_replicated4', ('fneg', a), b)),
+
# fdph works mostly like fdot, but to get the correct result, the negation
# must be applied to the second source.
(('fneg', ('fdph_replicated(is_used_once)', a, b)), ('fdph_replicated', a, ('fneg', b))),
- (('fabs', ('fdph_replicated(is_used_once)', a, b)), ('fdph_replicated', ('fabs', a), ('fabs', b))),
(('fneg', ('fsign(is_used_once)', a)), ('fsign', ('fneg', a))),
(('fabs', ('fsign(is_used_once)', a)), ('fsign', ('fabs', a))),
]
-for op in ['fmul', 'fdot_replicated2', 'fdot_replicated3', 'fdot_replicated4']:
- distribute_src_mods.extend([
- (('fneg', (op + '(is_used_once)', a, b)), (op, ('fneg', a), b)),
- (('fabs', (op + '(is_used_once)', a, b)), (op, ('fabs', a), ('fabs', b))),
- ])
-
print(nir_algebraic.AlgebraicPass("nir_opt_algebraic", optimizations).render())
print(nir_algebraic.AlgebraicPass("nir_opt_algebraic_before_ffma",
before_ffma_optimizations).render())
projects / mesa.git / blobdiff