# 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'),
+# Float sizes
+for s in [16, 32, 64]:
+ optimizations.extend([
+ (('~flrp@{}'.format(s), a, ('fadd', a, b), c), ('fadd', ('fmul', b, c), a), 'options->lower_flrp{}'.format(s)),
+ (('~flrp@{}'.format(s), ('fadd', a, b), ('fadd', a, c), d), ('fadd', ('flrp', b, c, d), a), 'options->lower_flrp{}'.format(s)),
+ (('~flrp@{}'.format(s), a, ('fmul(is_used_once)', a, b), c), ('fmul', ('flrp', 1.0, b, c), a), 'options->lower_flrp{}'.format(s)),
+
+ (('~flrp@{}'.format(s), a, b, ('b2f', 'c@1')), ('bcsel', c, b, a), 'options->lower_flrp{}'.format(s)),
+
+ (('~fadd@{}'.format(s), ('fmul', a, ('fadd', 1.0, ('fneg', c))), ('fmul', b, c)), ('flrp', a, b, c), '!options->lower_flrp{}'.format(s)),
+ # These are the same as the previous three rules, but it depends on
+ # 1-fsat(x) <=> fsat(1-x). See below.
+ (('~fadd@{}'.format(s), ('fmul', a, ('fsat', ('fadd', 1.0, ('fneg', c)))), ('fmul', b, ('fsat', c))), ('flrp', a, b, ('fsat', c)), '!options->lower_flrp{}'.format(s)),
+ (('~fadd@{}'.format(s), a, ('fmul', c, ('fadd', b, ('fneg', a)))), ('flrp', a, b, c), '!options->lower_flrp{}'.format(s)),
+
+ (('~fadd@{}'.format(s), ('fmul', a, ('fadd', 1.0, ('fneg', ('b2f', 'c@1')))), ('fmul', b, ('b2f', c))), ('bcsel', c, b, a), 'options->lower_flrp{}'.format(s)),
+ (('~fadd@{}'.format(s), a, ('fmul', ('b2f', 'c@1'), ('fadd', b, ('fneg', a)))), ('bcsel', c, b, a), 'options->lower_flrp{}'.format(s)),
+
+ # 1 - ((1 - a) * (1 - b))
+ # 1 - (1 - a - b + a*b)
+ # 1 - 1 + a + b - a*b
+ # a + b - a*b
+ # a + b*(1 - a)
+ # b*(1 - a) + 1*a
+ # flrp(b, 1, a)
+ (('~fadd@{}'.format(s), 1.0, ('fneg', ('fmul', ('fadd', 1.0, ('fneg', a)), ('fadd', 1.0, ('fneg', b))))), ('flrp', b, 1.0, a), '!options->lower_flrp{}'.format(s)),
+ ])
+optimizations.extend([
(('~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)),
(('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'),
(('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'),
- # These are the same as the previous three rules, but it depends on
- # 1-fsat(x) <=> fsat(1-x). See below.
- (('~fadd@32', ('fmul', a, ('fsat', ('fadd', 1.0, ('fneg', c )))), ('fmul', b, ('fsat', c))), ('flrp', a, b, ('fsat', c)), '!options->lower_flrp32'),
- (('~fadd@64', ('fmul', a, ('fsat', ('fadd', 1.0, ('fneg', c )))), ('fmul', b, ('fsat', c))), ('flrp', a, b, ('fsat', c)), '!options->lower_flrp64'),
-
- (('~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'),
# If x > 1: 1 - fsat(x) => 1 - 1 => 0 and fsat(1 - x) => fsat(< 0) => 0
(('~fadd', ('fneg(is_used_once)', ('fsat(is_used_once)', 'a(is_not_fmul)')), 1.0), ('fsat', ('fadd', 1.0, ('fneg', a)))),
- # 1 - ((1 - a) * (1 - b))
- # 1 - (1 - a - b + a*b)
- # 1 - 1 + a + b - a*b
- # a + b - a*b
- # a + b*(1 - a)
- # b*(1 - a) + 1*a
- # flrp(b, 1, a)
- (('~fadd@32', 1.0, ('fneg', ('fmul', ('fadd', 1.0, ('fneg', a)), ('fadd', 1.0, ('fneg', b))))),
- ('flrp', b, 1.0, a), '!options->lower_flrp32'),
-
# (a * #b + #c) << #d
# ((a * #b) << #d) + (#c << #d)
# (a * (#b << #d)) + (#c << #d)
# (a * #b) << #c
# a * (#b << #c)
(('ishl', ('imul', a, '#b'), '#c'), ('imul', a, ('ishl', b, c))),
-]
+])
# Care must be taken here. Shifts in NIR uses only the lower log2(bitsize)
# bits of the second source. These replacements must correctly handle the
(('~flrp', ('fadd(is_used_once)', a, b), ('fadd(is_used_once)', a, c), d), ('fadd', ('flrp', b, c, d), a)),
+ (('~fadd@16', 1.0, ('fmul(is_used_once)', c , ('fadd', b, -1.0 ))), ('fadd', ('fadd', 1.0, ('fneg', c)), ('fmul', b, c)), 'options->lower_flrp16'),
(('~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'),