-#! /usr/bin/env python
#
# Copyright (C) 2014 Intel Corporation
#
# however, be used for backend-requested lowering operations as those need to
# happen regardless of precision.
#
-# Variable names are specified as "[#]name[@type]" where "#" inicates that
-# the given variable will only match constants and the type indicates that
+# Variable names are specified as "[#]name[@type][(cond)]" where "#" inicates
+# that the given variable will only match constants and the type indicates that
# the given variable will only match values from ALU instructions with the
-# given output type.
+# given output type, and (cond) specifies an additional condition function
+# (see nir_search_helpers.h).
#
# For constants, you have to be careful to make sure that it is the right
# type because python is unaware of the source and destination types of the
# opcodes.
+#
+# 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.
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))))),
+ (('udiv', a, 1), a),
+ (('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))),
+ (('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'),
+ (('umod', a, '#b(is_pos_power_of_two)'), ('iand', a, ('isub', b, 1))),
+
(('fneg', ('fneg', a)), a),
(('ineg', ('ineg', a)), a),
(('fabs', ('fabs', a)), ('fabs', a)),
(('fabs', ('fneg', a)), ('fabs', a)),
+ (('fabs', ('u2f32', a)), ('u2f32', a)),
(('iabs', ('iabs', a)), ('iabs', a)),
(('iabs', ('ineg', a)), ('iabs', a)),
(('~fadd', a, 0.0), a),
(('~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_flrp'),
- (('flrp', a, b, c), ('fadd', ('fmul', c, ('fsub', b, a)), a), 'options->lower_flrp'),
+ (('~flrp', a, b, ('b2f', c)), ('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'),
(('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_flrp'),
- (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg', c ))), ('fmul', b, c )), ('flrp', a, b, c), '!options->lower_flrp'),
- (('~fadd', a, ('fmul', ('b2f', c), ('fadd', b, ('fneg', a)))), ('bcsel', c, b, a), 'options->lower_flrp'),
- (('~fadd', a, ('fmul', c , ('fadd', b, ('fneg', a)))), ('flrp', a, b, c), '!options->lower_flrp'),
+ (('~fadd', ('fmul', a, ('fadd', 1.0, ('fneg', ('b2f', c)))), ('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@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->lower_ffma'),
+ (('~fadd', ('fmul', a, b), c), ('ffma', a, b, c), 'options->fuse_ffma'),
+
+ # (a * #b + #c) << #d
+ # ((a * #b) << #d) + (#c << #d)
+ # (a * (#b << #d)) + (#c << #d)
+ (('ishl', ('iadd', ('imul', a, '#b'), '#c'), '#d'),
+ ('iadd', ('imul', a, ('ishl', b, d)), ('ishl', c, d))),
+
+ # (a * #b) << #c
+ # a * (#b << #c)
+ (('ishl', ('imul', a, '#b'), '#c'), ('imul', a, ('ishl', b, c))),
+
# Comparison simplifications
(('~inot', ('flt', a, b)), ('fge', a, b)),
(('~inot', ('fge', a, b)), ('flt', a, b)),
# inot(a)
(('fge', 0.0, ('b2f', a)), ('inot', a)),
+ (('fge', ('fneg', ('b2f', a)), 0.0), ('inot', 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)),
+ # 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)),
+
(('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', ('inot', 'a@bool'), b, c), ('bcsel', a, c, b)),
+ (('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)),
(('fmin', a, a), a),
(('fmax', a, a), a),
(('imin', a, a), a),
(('imax', a, a), a),
(('umin', a, a), a),
(('umax', a, a), a),
+ (('fmin', a, ('fneg', a)), ('fneg', ('fabs', a))),
+ (('imin', a, ('ineg', a)), ('ineg', ('iabs', a))),
+ (('fmin', a, ('fneg', ('fabs', a))), ('fneg', ('fabs', a))),
+ (('imin', a, ('ineg', ('iabs', a))), ('ineg', ('iabs', a))),
+ (('fmin', a, ('fabs', a)), a),
+ (('imin', a, ('iabs', a)), a),
+ (('fmax', a, ('fneg', ('fabs', a))), a),
+ (('imax', a, ('ineg', ('iabs', a))), a),
+ (('fmax', a, ('fabs', a)), ('fabs', a)),
+ (('imax', a, ('iabs', a)), ('iabs', a)),
+ (('fmax', a, ('fneg', a)), ('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', a), ('fmin', ('fmax', a, 0.0), 1.0), 'options->lower_fsat'),
(('fsat', ('fsat', a)), ('fsat', a)),
- (('fmin', ('fmax', ('fmin', ('fmax', a, 0.0), 1.0), 0.0), 1.0), ('fmin', ('fmax', a, 0.0), 1.0)),
+ (('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))),
+ (('extract_u8', ('imin', ('imax', a, 0), 0xff), 0), ('imin', ('imax', a, 0), 0xff)),
(('~ior', ('flt', a, b), ('flt', a, c)), ('flt', a, ('fmax', b, c))),
(('~ior', ('flt', a, c), ('flt', b, c)), ('flt', ('fmin', a, b), c)),
(('~ior', ('fge', a, b), ('fge', a, c)), ('fge', a, ('fmin', b, c))),
(('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)),
+ # 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.
# Comparison with the same args. Note that these are not done for
(('iand', a, 0), 0),
(('ior', a, a), a),
(('ior', a, 0), a),
+ (('ior', a, True), True),
(('fxor', a, a), 0.0),
(('ixor', a, a), 0),
(('ixor', a, 0), a),
(('~flog2', ('frcp', a)), ('fneg', ('flog2', a))),
(('~flog2', ('frsq', a)), ('fmul', -0.5, ('flog2', a))),
(('~flog2', ('fpow', a, b)), ('fmul', b, ('flog2', a))),
- (('~fadd', ('flog2', a), ('flog2', b)), ('flog2', ('fmul', a, b))),
- (('~fadd', ('flog2', a), ('fneg', ('flog2', b))), ('flog2', ('fdiv', a, b))),
(('~fmul', ('fexp2', a), ('fexp2', b)), ('fexp2', ('fadd', a, b))),
# Division and reciprocal
(('~fdiv', 1.0, a), ('frcp', a)),
(('~frcp', ('frsq', a)), ('fsqrt', a), '!options->lower_fsqrt'),
# Boolean simplifications
(('ieq', 'a@bool', True), a),
- (('ine', 'a@bool', True), ('inot', a)),
+ (('ine(is_not_used_by_if)', 'a@bool', True), ('inot', a)),
(('ine', 'a@bool', False), a),
- (('ieq', 'a@bool', False), ('inot', 'a')),
- (('bcsel', a, True, False), ('ine', a, 0)),
- (('bcsel', a, False, True), ('ieq', a, 0)),
+ (('ieq(is_not_used_by_if)', 'a@bool', 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', True, b, c), b),
(('bcsel', False, b, c), c),
# The result of this should be hit by constant propagation and, in the
# Conversions
(('i2b', ('b2i', a)), a),
- (('f2i', ('ftrunc', a)), ('f2i', a)),
- (('f2u', ('ftrunc', a)), ('f2u', a)),
+ (('f2i32', ('ftrunc', a)), ('f2i32', a)),
+ (('f2u32', ('ftrunc', a)), ('f2u32', a)),
(('i2b', ('ineg', a)), ('i2b', a)),
(('i2b', ('iabs', a)), ('i2b', a)),
(('fabs', ('b2f', a)), ('b2f', a)),
(('iabs', ('b2i', a)), ('b2i', a)),
+ # 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),
+ (('pack_64_2x32_split', ('unpack_64_2x32_split_x', 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'),
# Propagate negation up multiplication chains
(('fmul', ('fneg', a), b), ('fneg', ('fmul', a, b))),
- (('imul', ('ineg', a), b), ('ineg', ('fmul', a, b))),
+ (('imul', ('ineg', a), b), ('ineg', ('imul', a, b))),
+
+ # Propagate constants up multiplication chains
+ (('~fmul(is_used_once)', ('fmul(is_used_once)', 'a(is_not_const)', 'b(is_not_const)'), '#c'), ('fmul', ('fmul', a, c), b)),
+ (('imul(is_used_once)', ('imul(is_used_once)', 'a(is_not_const)', 'b(is_not_const)'), '#c'), ('imul', ('imul', a, c), b)),
+ (('~fadd(is_used_once)', ('fadd(is_used_once)', 'a(is_not_const)', 'b(is_not_const)'), '#c'), ('fadd', ('fadd', a, c), b)),
+ (('iadd(is_used_once)', ('iadd(is_used_once)', 'a(is_not_const)', 'b(is_not_const)'), '#c'), ('iadd', ('iadd', a, c), b)),
+
+ # 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)),
+ (('iadd', '#a', ('iadd', b, '#c')), ('iadd', ('iadd', a, c), b)),
# Misc. lowering
- (('fmod', a, b), ('fsub', a, ('fmul', b, ('ffloor', ('fdiv', a, b)))), 'options->lower_fmod'),
- (('uadd_carry', a, b), ('b2i', ('ult', ('iadd', a, b), a)), 'options->lower_uadd_carry'),
- (('usub_borrow', a, b), ('b2i', ('ult', a, b)), 'options->lower_usub_borrow'),
+ (('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'),
+ (('uadd_carry@32', 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'),
('bcsel', ('ilt', 31, 'bits'), 'insert',
('ubfe', 'value', 'offset', 'bits')),
'options->lower_bitfield_extract'),
- (('extract_i8', a, b),
+ (('extract_i8', a, 'b@32'),
('ishr', ('ishl', a, ('imul', ('isub', 3, b), 8)), 24),
'options->lower_extract_byte'),
- (('extract_u8', a, b),
+ (('extract_u8', a, 'b@32'),
('iand', ('ushr', a, ('imul', b, 8)), 0xff),
'options->lower_extract_byte'),
- (('extract_i16', a, b),
+ (('extract_i16', a, 'b@32'),
('ishr', ('ishl', a, ('imul', ('isub', 1, b), 16)), 16),
'options->lower_extract_word'),
- (('extract_u16', a, b),
+ (('extract_u16', a, 'b@32'),
('iand', ('ushr', a, ('imul', b, 16)), 0xffff),
'options->lower_extract_word'),
(('pack_unorm_2x16', 'v'),
('pack_uvec2_to_uint',
- ('f2u', ('fround_even', ('fmul', ('fsat', 'v'), 65535.0)))),
+ ('f2u32', ('fround_even', ('fmul', ('fsat', 'v'), 65535.0)))),
'options->lower_pack_unorm_2x16'),
(('pack_unorm_4x8', 'v'),
('pack_uvec4_to_uint',
- ('f2u', ('fround_even', ('fmul', ('fsat', 'v'), 255.0)))),
+ ('f2u32', ('fround_even', ('fmul', ('fsat', 'v'), 255.0)))),
'options->lower_pack_unorm_4x8'),
(('pack_snorm_2x16', 'v'),
('pack_uvec2_to_uint',
- ('f2i', ('fround_even', ('fmul', ('fmin', 1.0, ('fmax', -1.0, 'v')), 32767.0)))),
+ ('f2i32', ('fround_even', ('fmul', ('fmin', 1.0, ('fmax', -1.0, 'v')), 32767.0)))),
'options->lower_pack_snorm_2x16'),
(('pack_snorm_4x8', 'v'),
('pack_uvec4_to_uint',
- ('f2i', ('fround_even', ('fmul', ('fmin', 1.0, ('fmax', -1.0, 'v')), 127.0)))),
+ ('f2i32', ('fround_even', ('fmul', ('fmin', 1.0, ('fmax', -1.0, 'v')), 127.0)))),
'options->lower_pack_snorm_4x8'),
(('unpack_unorm_2x16', 'v'),
- ('fdiv', ('u2f', ('vec2', ('extract_u16', 'v', 0),
- ('extract_u16', 'v', 1))),
+ ('fdiv', ('u2f32', ('vec2', ('extract_u16', 'v', 0),
+ ('extract_u16', 'v', 1))),
65535.0),
'options->lower_unpack_unorm_2x16'),
(('unpack_unorm_4x8', 'v'),
- ('fdiv', ('u2f', ('vec4', ('extract_u8', 'v', 0),
- ('extract_u8', 'v', 1),
- ('extract_u8', 'v', 2),
- ('extract_u8', 'v', 3))),
+ ('fdiv', ('u2f32', ('vec4', ('extract_u8', 'v', 0),
+ ('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', ('i2f', ('vec2', ('extract_i16', 'v', 0),
- ('extract_i16', 'v', 1))),
+ ('fmin', 1.0, ('fmax', -1.0, ('fdiv', ('i2f32', ('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', ('i2f', ('vec4', ('extract_i8', 'v', 0),
- ('extract_i8', 'v', 1),
- ('extract_i8', 'v', 2),
- ('extract_i8', 'v', 3))),
+ ('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))),
127.0))),
'options->lower_unpack_snorm_4x8'),
]
+def fexp2i(exp, bits):
+ # We assume that exp is already in the right range.
+ if bits == 32:
+ return ('ishl', ('iadd', exp, 127), 23)
+ elif bits == 64:
+ return ('pack_64_2x32_split', 0, ('ishl', ('iadd', exp, 1023), 20))
+ else:
+ assert False
+
+def ldexp(f, exp, bits):
+ # First, we clamp exp to a reasonable range. The maximum possible range
+ # for a normal exponent is [-126, 127] and, throwing in denormals, you get
+ # a maximum range of [-149, 127]. This means that we can potentially have
+ # a swing of +-276. If you start with FLT_MAX, you actually have to do
+ # ldexp(FLT_MAX, -278) to get it to flush all the way to zero. The GLSL
+ # spec, on the other hand, only requires that we handle an exponent value
+ # in the range [-126, 128]. This implementation is *mostly* correct; it
+ # 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:
+ exp = ('imin', ('imax', exp, -252), 254)
+ elif bits == 64:
+ exp = ('imin', ('imax', exp, -2044), 2046)
+ else:
+ assert False
+
+ # Now we compute two powers of 2, one for exp/2 and one for exp-exp/2.
+ # (We use ishr which isn't the same for -1, but the -1 case still works
+ # since we use exp-exp/2 as the second exponent.) While the spec
+ # technically defines ldexp as f * 2.0^exp, simply multiplying once doesn't
+ # work with denormals and doesn't allow for the full swing in exponents
+ # that you can get with normalized values. Instead, we create two powers
+ # of two and multiply by them each in turn. That way the effective range
+ # of our exponent is doubled.
+ pow2_1 = fexp2i(('ishr', exp, 1), bits)
+ pow2_2 = fexp2i(('isub', exp, ('ishr', exp, 1)), bits)
+ return ('fmul', ('fmul', f, pow2_1), pow2_2)
+
+optimizations += [
+ (('ldexp@32', 'x', 'exp'), ldexp('x', 'exp', 32)),
+ (('ldexp@64', 'x', 'exp'), ldexp('x', 'exp', 64)),
+]
+
# Unreal Engine 4 demo applications open-codes bitfieldReverse()
def bitfield_reverse(u):
step1 = ('ior', ('ishl', u, 16), ('ushr', u, 16))
return step5
-optimizations += [(bitfield_reverse('x'), ('bitfield_reverse', 'x'))]
+optimizations += [(bitfield_reverse('x@32'), ('bitfield_reverse', 'x'))]
+# For any float comparison operation, "cmp", if you have "a == a && a cmp b"
+# then the "a == a" is redundant because it's equivalent to "a is not NaN"
+# and, if a is a NaN then the second comparison will fail anyway.
+for op in ['flt', 'fge', 'feq']:
+ optimizations += [
+ (('iand', ('feq', a, a), (op, a, b)), (op, a, b)),
+ (('iand', ('feq', a, a), (op, b, a)), (op, b, a)),
+ ]
# Add optimizations to handle the case where the result of a ternary is
# compared to a constant. This way we can take things like
('bcsel', 'a', (op, 'd', 'b'), (op, 'd', 'c'))),
]
+# 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
+# with the regular optimizations.
+before_ffma_optimizations = [
+ # Propagate constants down multiplication chains
+ (('~fmul(is_used_once)', ('fmul(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('fmul', ('fmul', a, c), b)),
+ (('imul(is_used_once)', ('imul(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('imul', ('imul', a, c), b)),
+ (('~fadd(is_used_once)', ('fadd(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('fadd', ('fadd', a, c), b)),
+ (('iadd(is_used_once)', ('iadd(is_used_once)', 'a(is_not_const)', '#b'), 'c(is_not_const)'), ('iadd', ('iadd', a, c), b)),
+
+ (('~fadd', ('fmul', a, b), ('fmul', a, c)), ('fmul', a, ('fadd', b, c))),
+ (('iadd', ('imul', a, b), ('imul', a, c)), ('imul', a, ('iadd', b, c))),
+ (('~fadd', ('fneg', a), a), 0.0),
+ (('iadd', ('ineg', a), a), 0),
+ (('iadd', ('ineg', a), ('iadd', a, b)), b),
+ (('iadd', a, ('iadd', ('ineg', a), b)), b),
+ (('~fadd', ('fneg', a), ('fadd', a, b)), b),
+ (('~fadd', a, ('fadd', ('fneg', a), b)), b),
+]
+
# This section contains "late" optimizations that should be run after the
# regular optimizations have finished. Optimizations should go here if
# they help code generation but do not necessarily produce code that is
(('fdot3', a, b), ('fdot_replicated3', a, b), 'options->fdot_replicates'),
(('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)),
+
+ # 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))),
]
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()