2014-12-12 19:13:10 +00:00
|
|
|
#! /usr/bin/env python
|
|
|
|
#
|
|
|
|
# Copyright (C) 2014 Intel Corporation
|
|
|
|
#
|
|
|
|
# Permission is hereby granted, free of charge, to any person obtaining a
|
|
|
|
# copy of this software and associated documentation files (the "Software"),
|
|
|
|
# to deal in the Software without restriction, including without limitation
|
|
|
|
# the rights to use, copy, modify, merge, publish, distribute, sublicense,
|
|
|
|
# and/or sell copies of the Software, and to permit persons to whom the
|
|
|
|
# Software is furnished to do so, subject to the following conditions:
|
|
|
|
#
|
|
|
|
# The above copyright notice and this permission notice (including the next
|
|
|
|
# paragraph) shall be included in all copies or substantial portions of the
|
|
|
|
# Software.
|
|
|
|
#
|
|
|
|
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
|
|
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
|
|
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
|
|
|
|
# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
|
|
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
|
|
|
|
# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
|
|
|
|
# IN THE SOFTWARE.
|
|
|
|
#
|
|
|
|
# Authors:
|
|
|
|
# Jason Ekstrand (jason@jlekstrand.net)
|
|
|
|
|
|
|
|
import nir_algebraic
|
|
|
|
|
|
|
|
# Convenience variables
|
|
|
|
a = 'a'
|
|
|
|
b = 'b'
|
|
|
|
c = 'c'
|
|
|
|
d = 'd'
|
|
|
|
|
|
|
|
# Written in the form (<search>, <replace>) where <search> is an expression
|
|
|
|
# and <replace> is either an expression or a value. An expression is
|
2016-03-17 18:04:49 +00:00
|
|
|
# defined as a tuple of the form ([~]<op>, <src0>, <src1>, <src2>, <src3>)
|
2014-12-12 19:13:10 +00:00
|
|
|
# where each source is either an expression or a value. A value can be
|
2015-01-29 00:42:20 +00:00
|
|
|
# either a numeric constant or a string representing a variable name.
|
|
|
|
#
|
2016-03-17 18:04:49 +00:00
|
|
|
# If the opcode in a search expression is prefixed by a '~' character, this
|
|
|
|
# indicates that the operation is inexact. Such operations will only get
|
|
|
|
# applied to SSA values that do not have the exact bit set. This should be
|
|
|
|
# used by by any optimizations that are not bit-for-bit exact. It should not,
|
|
|
|
# however, be used for backend-requested lowering operations as those need to
|
|
|
|
# happen regardless of precision.
|
|
|
|
#
|
2015-01-29 00:42:20 +00:00
|
|
|
# Variable names are specified as "[#]name[@type]" 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.
|
|
|
|
#
|
|
|
|
# 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
|
2014-12-12 19:13:10 +00:00
|
|
|
# opcodes.
|
|
|
|
|
|
|
|
optimizations = [
|
|
|
|
(('fneg', ('fneg', a)), a),
|
|
|
|
(('ineg', ('ineg', a)), a),
|
|
|
|
(('fabs', ('fabs', a)), ('fabs', a)),
|
|
|
|
(('fabs', ('fneg', a)), ('fabs', a)),
|
|
|
|
(('iabs', ('iabs', a)), ('iabs', a)),
|
|
|
|
(('iabs', ('ineg', a)), ('iabs', a)),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fadd', a, 0.0), a),
|
2014-12-12 19:13:10 +00:00
|
|
|
(('iadd', a, 0), a),
|
2015-08-19 06:38:34 +01:00
|
|
|
(('usadd_4x8', a, 0), a),
|
|
|
|
(('usadd_4x8', a, ~0), ~0),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fadd', ('fmul', a, b), ('fmul', a, c)), ('fmul', a, ('fadd', b, c))),
|
2015-02-28 19:32:31 +00:00
|
|
|
(('iadd', ('imul', a, b), ('imul', a, c)), ('imul', a, ('iadd', b, c))),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fadd', ('fneg', a), a), 0.0),
|
2015-02-28 19:32:32 +00:00
|
|
|
(('iadd', ('ineg', a), a), 0),
|
2015-12-15 01:44:23 +00:00
|
|
|
(('iadd', ('ineg', a), ('iadd', a, b)), b),
|
|
|
|
(('iadd', a, ('iadd', ('ineg', a), b)), b),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fadd', ('fneg', a), ('fadd', a, b)), b),
|
|
|
|
(('~fadd', a, ('fadd', ('fneg', a), b)), b),
|
|
|
|
(('~fmul', a, 0.0), 0.0),
|
2014-12-12 19:13:10 +00:00
|
|
|
(('imul', a, 0), 0),
|
2015-08-19 06:38:34 +01:00
|
|
|
(('umul_unorm_4x8', a, 0), 0),
|
|
|
|
(('umul_unorm_4x8', a, ~0), a),
|
2014-12-12 19:13:10 +00:00
|
|
|
(('fmul', a, 1.0), a),
|
|
|
|
(('imul', a, 1), a),
|
|
|
|
(('fmul', a, -1.0), ('fneg', a)),
|
|
|
|
(('imul', a, -1), ('ineg', a)),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~ffma', 0.0, a, b), b),
|
|
|
|
(('~ffma', a, 0.0, b), b),
|
|
|
|
(('~ffma', a, b, 0.0), ('fmul', a, b)),
|
2014-12-12 19:13:10 +00:00
|
|
|
(('ffma', a, 1.0, b), ('fadd', a, b)),
|
|
|
|
(('ffma', 1.0, a, b), ('fadd', a, b)),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~flrp', a, b, 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'),
|
2015-02-18 20:24:38 +00:00
|
|
|
(('flrp', a, b, c), ('fadd', ('fmul', c, ('fsub', b, a)), a), 'options->lower_flrp'),
|
2015-09-14 16:13:19 +01:00
|
|
|
(('ffract', a), ('fsub', a, ('ffloor', a)), 'options->lower_ffract'),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~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'),
|
2015-01-30 21:49:48 +00:00
|
|
|
(('ffma', a, b, c), ('fadd', ('fmul', a, b), c), 'options->lower_ffma'),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fadd', ('fmul', a, b), c), ('ffma', a, b, c), '!options->lower_ffma'),
|
2015-01-22 07:36:01 +00:00
|
|
|
# Comparison simplifications
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~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)),
|
2015-01-22 07:36:01 +00:00
|
|
|
(('inot', ('ilt', a, b)), ('ige', a, b)),
|
|
|
|
(('inot', ('ige', a, b)), ('ilt', a, b)),
|
2015-05-06 06:54:59 +01:00
|
|
|
(('inot', ('ieq', a, b)), ('ine', a, b)),
|
|
|
|
(('inot', ('ine', a, b)), ('ieq', a, b)),
|
2016-03-02 23:18:34 +00:00
|
|
|
|
|
|
|
# 0.0 >= b2f(a)
|
|
|
|
# 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)),
|
|
|
|
|
2016-03-02 23:36:14 +00:00
|
|
|
# 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)),
|
|
|
|
|
2014-12-12 19:13:10 +00:00
|
|
|
(('fge', ('fneg', ('fabs', a)), 0.0), ('feq', a, 0.0)),
|
2016-03-23 21:30:29 +00:00
|
|
|
(('bcsel', ('flt', b, a), b, a), ('fmin', a, b)),
|
2015-02-11 05:43:32 +00:00
|
|
|
(('bcsel', ('flt', a, b), b, a), ('fmax', a, b)),
|
2015-02-20 09:18:46 +00:00
|
|
|
(('bcsel', ('inot', 'a@bool'), b, c), ('bcsel', a, c, b)),
|
2015-02-20 09:20:34 +00:00
|
|
|
(('bcsel', a, ('bcsel', a, b, c), d), ('bcsel', a, b, d)),
|
nir: Simplify a bcsel to logical-or
Oddly, this did not affect the shader where I first noticed the pattern.
That particular shader doesn't get its if-statement converted to a bcsel
because there are two assignments in the else-statement. This led to me
submitting https://bugs.freedesktop.org/show_bug.cgi?id=94747.
shader-db results:
Sandy Bridge
total instructions in shared programs: 8467384 -> 8467069 (-0.00%)
instructions in affected programs: 36594 -> 36279 (-0.86%)
helped: 46
HURT: 0
total cycles in shared programs: 117573448 -> 117568518 (-0.00%)
cycles in affected programs: 339114 -> 334184 (-1.45%)
helped: 46
HURT: 0
Ivy Bridge / Haswell / Broadwell / Skylake:
total instructions in shared programs: 7774258 -> 7773999 (-0.00%)
instructions in affected programs: 30874 -> 30615 (-0.84%)
helped: 46
HURT: 0
total cycles in shared programs: 65739190 -> 65734530 (-0.01%)
cycles in affected programs: 180380 -> 175720 (-2.58%)
helped: 45
HURT: 1
No change on G45 or Ironlake.
I also tried these expressions, but none of them affected any shaders in
shader-db:
(('bcsel', a, 'a@bool', 'b@bool'), ('ior', a, b)),
(('bcsel', a, 'b@bool', False), ('iand', a, b)),
(('bcsel', a, 'b@bool', 'a@bool'), ('iand', a, b)),
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
2016-03-29 23:25:04 +01:00
|
|
|
(('bcsel', a, True, 'b@bool'), ('ior', a, b)),
|
2015-05-06 04:20:30 +01:00
|
|
|
(('fmin', a, a), a),
|
|
|
|
(('fmax', a, a), a),
|
|
|
|
(('imin', a, a), a),
|
|
|
|
(('imax', a, a), a),
|
|
|
|
(('umin', a, a), a),
|
|
|
|
(('umax', a, a), a),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fmin', ('fmax', a, 0.0), 1.0), ('fsat', a), '!options->lower_fsat'),
|
|
|
|
(('~fmax', ('fmin', a, 1.0), 0.0), ('fsat', a), '!options->lower_fsat'),
|
2015-01-30 21:53:39 +00:00
|
|
|
(('fsat', a), ('fmin', ('fmax', a, 0.0), 1.0), 'options->lower_fsat'),
|
2015-01-30 22:10:58 +00:00
|
|
|
(('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)),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~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))),
|
|
|
|
(('~ior', ('fge', a, c), ('fge', b, c)), ('fge', ('fmax', a, b), c)),
|
2016-03-02 21:47:56 +00:00
|
|
|
(('fabs', ('slt', a, b)), ('slt', a, b)),
|
|
|
|
(('fabs', ('sge', a, b)), ('sge', a, b)),
|
|
|
|
(('fabs', ('seq', a, b)), ('seq', a, b)),
|
|
|
|
(('fabs', ('sne', a, b)), ('sne', a, b)),
|
2015-03-31 16:25:19 +01:00
|
|
|
(('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'),
|
2015-08-06 12:36:04 +01:00
|
|
|
(('fne', ('fneg', a), a), ('fne', a, 0.0)),
|
2015-08-06 12:36:05 +01:00
|
|
|
(('feq', ('fneg', a), a), ('feq', a, 0.0)),
|
2015-03-18 06:30:32 +00:00
|
|
|
# Emulating booleans
|
2015-05-06 03:50:58 +01:00
|
|
|
(('imul', ('b2i', a), ('b2i', b)), ('b2i', ('iand', a, b))),
|
2015-03-18 06:30:32 +00:00
|
|
|
(('fmul', ('b2f', a), ('b2f', b)), ('b2f', ('iand', a, b))),
|
2015-03-18 06:30:43 +00:00
|
|
|
(('fsat', ('fadd', ('b2f', a), ('b2f', b))), ('b2f', ('ior', a, b))),
|
2015-03-30 06:59:39 +01:00
|
|
|
(('iand', 'a@bool', 1.0), ('b2f', a)),
|
2015-03-30 07:04:21 +01:00
|
|
|
(('flt', ('fneg', ('b2f', a)), 0), a), # Generated by TGSI KILL_IF.
|
|
|
|
(('flt', ('fsub', 0.0, ('b2f', a)), 0), a), # Generated by TGSI KILL_IF.
|
2015-02-07 01:12:59 +00:00
|
|
|
# Comparison with the same args. Note that these are not done for
|
|
|
|
# the float versions because NaN always returns false on float
|
|
|
|
# inequalities.
|
|
|
|
(('ilt', a, a), False),
|
|
|
|
(('ige', a, a), True),
|
|
|
|
(('ieq', a, a), True),
|
|
|
|
(('ine', a, a), False),
|
|
|
|
(('ult', a, a), False),
|
|
|
|
(('uge', a, a), True),
|
2015-01-19 22:57:38 +00:00
|
|
|
# Logical and bit operations
|
|
|
|
(('fand', a, 0.0), 0.0),
|
|
|
|
(('iand', a, a), a),
|
2015-07-31 17:12:48 +01:00
|
|
|
(('iand', a, ~0), a),
|
2015-01-19 22:57:38 +00:00
|
|
|
(('iand', a, 0), 0),
|
|
|
|
(('ior', a, a), a),
|
|
|
|
(('ior', a, 0), a),
|
|
|
|
(('fxor', a, a), 0.0),
|
|
|
|
(('ixor', a, a), 0),
|
2016-02-02 00:34:46 +00:00
|
|
|
(('ixor', a, 0), a),
|
2015-01-19 22:57:38 +00:00
|
|
|
(('inot', ('inot', a)), a),
|
|
|
|
# DeMorgan's Laws
|
|
|
|
(('iand', ('inot', a), ('inot', b)), ('inot', ('ior', a, b))),
|
|
|
|
(('ior', ('inot', a), ('inot', b)), ('inot', ('iand', a, b))),
|
2015-01-22 07:25:56 +00:00
|
|
|
# Shift optimizations
|
|
|
|
(('ishl', 0, a), 0),
|
|
|
|
(('ishl', a, 0), a),
|
|
|
|
(('ishr', 0, a), 0),
|
|
|
|
(('ishr', a, 0), a),
|
|
|
|
(('ushr', 0, a), 0),
|
2015-04-14 16:40:22 +01:00
|
|
|
(('ushr', a, 0), a),
|
nir: Recognize open-coded extract_u8.
Two shaders that appear in Unigine benchmarks (Heaven and Valley) unpack
three bytes from an integer and convert each into a float:
float((val >> 16u) & 0xffu)
float((val >> 8u) & 0xffu)
float((val >> 0u) & 0xffu)
Instead of shifting, masking, and type converting like this:
shr(8) g15<1>UD g25<8,8,1>UD 0x00000010UD
and(8) g16<1>UD g15<8,8,1>UD 0x000000ffUD
mov(8) g17<1>F g16<8,8,1>UD
shr(8) g18<1>UD g25<8,8,1>UD 0x00000008UD
and(8) g19<1>UD g18<8,8,1>UD 0x000000ffUD
mov(8) g20<1>F g19<8,8,1>UD
and(8) g21<1>UD g25<8,8,1>UD 0x000000ffUD
mov(8) g22<1>F g21<8,8,1>UD
i965 can simply extract a byte and convert to float in a single
instruction:
mov(8) g17<1>F g25.2<32,8,4>UB
mov(8) g20<1>F g25.1<32,8,4>UB
mov(8) g22<1>F g25.0<32,8,4>UB
This patch implements the first step: recognizing byte extraction. A
later patch will optimize out the conversion to float.
instructions in affected programs: 28568 -> 27450 (-3.91%)
helped: 7
cycles in affected programs: 210076 -> 203144 (-3.30%)
helped: 7
This patch decreases the number of instructions in the two Unigine
programs by:
#1721: 4520 -> 4374 instructions (-3.23%)
#1706: 3752 -> 3582 instructions (-4.53%)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
2016-01-21 17:09:48 +00:00
|
|
|
(('iand', 0xff, ('ushr', a, 24)), ('ushr', a, 24)),
|
2016-02-03 22:28:31 +00:00
|
|
|
(('iand', 0xffff, ('ushr', a, 16)), ('ushr', a, 16)),
|
2015-01-22 07:47:06 +00:00
|
|
|
# Exponential/logarithmic identities
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fexp2', ('flog2', a)), a), # 2^lg2(a) = a
|
|
|
|
(('~flog2', ('fexp2', a)), a), # lg2(2^a) = a
|
2015-01-28 18:32:57 +00:00
|
|
|
(('fpow', a, b), ('fexp2', ('fmul', ('flog2', a), b)), 'options->lower_fpow'), # a^b = 2^(lg2(a)*b)
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fexp2', ('fmul', ('flog2', a), b)), ('fpow', a, b), '!options->lower_fpow'), # 2^(lg2(a)*b) = a^b
|
|
|
|
(('~fexp2', ('fadd', ('fmul', ('flog2', a), b), ('fmul', ('flog2', c), d))),
|
|
|
|
('~fmul', ('fpow', a, b), ('fpow', c, d)), '!options->lower_fpow'), # 2^(lg2(a) * b + lg2(c) + d) = a^b * c^d
|
|
|
|
(('~fpow', a, 1.0), a),
|
|
|
|
(('~fpow', a, 2.0), ('fmul', a, a)),
|
|
|
|
(('~fpow', a, 4.0), ('fmul', ('fmul', a, a), ('fmul', a, a))),
|
|
|
|
(('~fpow', 2.0, a), ('fexp2', a)),
|
|
|
|
(('~fpow', ('fpow', a, 2.2), 0.454545), a),
|
|
|
|
(('~fpow', ('fabs', ('fpow', a, 2.2)), 0.454545), ('fabs', a)),
|
|
|
|
(('~fsqrt', ('fexp2', a)), ('fexp2', ('fmul', 0.5, a))),
|
|
|
|
(('~frcp', ('fexp2', a)), ('fexp2', ('fneg', a))),
|
|
|
|
(('~frsq', ('fexp2', a)), ('fexp2', ('fmul', -0.5, a))),
|
|
|
|
(('~flog2', ('fsqrt', a)), ('fmul', 0.5, ('flog2', 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))),
|
2015-01-22 07:52:17 +00:00
|
|
|
# Division and reciprocal
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fdiv', 1.0, a), ('frcp', a)),
|
nir: Add a lower_fdiv option, turn fdiv into fmul/frcp.
The nir_opt_algebraic rule
(('fadd', ('flog2', a), ('fneg', ('flog2', b))), ('flog2', ('fdiv', a, b))),
can produce new fdiv operations, which need to be lowered on i965,
as we don't actually implement fdiv. (Normally, we handle this in
GLSL IR's lower_instructions pass, but in the above case we introduce
an fdiv after that point. So, make NIR do it for us.)
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason.ekstrand@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Cc: mesa-stable@lists.freedesktop.org
2016-01-05 13:09:46 +00:00
|
|
|
(('fdiv', a, b), ('fmul', a, ('frcp', b)), 'options->lower_fdiv'),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~frcp', ('frcp', a)), a),
|
|
|
|
(('~frcp', ('fsqrt', a)), ('frsq', a)),
|
2015-01-28 18:39:29 +00:00
|
|
|
(('fsqrt', a), ('frcp', ('frsq', a)), 'options->lower_fsqrt'),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~frcp', ('frsq', a)), ('fsqrt', a), '!options->lower_fsqrt'),
|
2015-01-29 00:55:03 +00:00
|
|
|
# Boolean simplifications
|
2015-12-02 00:13:11 +00:00
|
|
|
(('ieq', 'a@bool', True), a),
|
|
|
|
(('ine', 'a@bool', True), ('inot', a)),
|
|
|
|
(('ine', 'a@bool', False), a),
|
|
|
|
(('ieq', 'a@bool', False), ('inot', 'a')),
|
2015-01-29 00:53:51 +00:00
|
|
|
(('bcsel', a, True, False), ('ine', a, 0)),
|
|
|
|
(('bcsel', a, False, True), ('ieq', a, 0)),
|
|
|
|
(('bcsel', True, b, c), b),
|
|
|
|
(('bcsel', False, b, c), 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)),
|
2015-01-19 22:57:38 +00:00
|
|
|
|
2015-01-29 23:50:18 +00:00
|
|
|
(('bcsel', a, b, b), b),
|
|
|
|
(('fcsel', a, b, b), b),
|
|
|
|
|
2015-03-26 17:09:42 +00:00
|
|
|
# Conversions
|
2015-05-06 03:56:27 +01:00
|
|
|
(('i2b', ('b2i', a)), a),
|
2015-03-26 17:09:42 +00:00
|
|
|
(('f2i', ('ftrunc', a)), ('f2i', a)),
|
|
|
|
(('f2u', ('ftrunc', a)), ('f2u', a)),
|
2016-03-02 02:59:57 +00:00
|
|
|
(('i2b', ('ineg', a)), ('i2b', a)),
|
|
|
|
(('i2b', ('iabs', a)), ('i2b', a)),
|
2016-03-02 21:46:50 +00:00
|
|
|
(('fabs', ('b2f', a)), ('b2f', a)),
|
|
|
|
(('iabs', ('b2i', a)), ('b2i', a)),
|
2015-03-26 17:09:42 +00:00
|
|
|
|
nir: Recognize open-coded extract_u8.
Two shaders that appear in Unigine benchmarks (Heaven and Valley) unpack
three bytes from an integer and convert each into a float:
float((val >> 16u) & 0xffu)
float((val >> 8u) & 0xffu)
float((val >> 0u) & 0xffu)
Instead of shifting, masking, and type converting like this:
shr(8) g15<1>UD g25<8,8,1>UD 0x00000010UD
and(8) g16<1>UD g15<8,8,1>UD 0x000000ffUD
mov(8) g17<1>F g16<8,8,1>UD
shr(8) g18<1>UD g25<8,8,1>UD 0x00000008UD
and(8) g19<1>UD g18<8,8,1>UD 0x000000ffUD
mov(8) g20<1>F g19<8,8,1>UD
and(8) g21<1>UD g25<8,8,1>UD 0x000000ffUD
mov(8) g22<1>F g21<8,8,1>UD
i965 can simply extract a byte and convert to float in a single
instruction:
mov(8) g17<1>F g25.2<32,8,4>UB
mov(8) g20<1>F g25.1<32,8,4>UB
mov(8) g22<1>F g25.0<32,8,4>UB
This patch implements the first step: recognizing byte extraction. A
later patch will optimize out the conversion to float.
instructions in affected programs: 28568 -> 27450 (-3.91%)
helped: 7
cycles in affected programs: 210076 -> 203144 (-3.30%)
helped: 7
This patch decreases the number of instructions in the two Unigine
programs by:
#1721: 4520 -> 4374 instructions (-3.23%)
#1706: 3752 -> 3582 instructions (-4.53%)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
2016-01-21 17:09:48 +00:00
|
|
|
# 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'),
|
|
|
|
|
2016-02-03 22:28:31 +00:00
|
|
|
# Word extraction
|
|
|
|
(('ushr', a, 16), ('extract_u16', a, 1), '!options->lower_extract_word'),
|
|
|
|
(('iand', 0xffff, a), ('extract_u16', a, 0), '!options->lower_extract_word'),
|
|
|
|
|
2015-01-28 18:52:53 +00:00
|
|
|
# Subtracts
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fsub', a, ('fsub', 0.0, b)), ('fadd', a, b)),
|
2015-02-20 08:00:27 +00:00
|
|
|
(('isub', a, ('isub', 0, b)), ('iadd', a, b)),
|
2015-08-19 06:38:34 +01:00
|
|
|
(('ussub_4x8', a, 0), a),
|
|
|
|
(('ussub_4x8', a, ~0), 0),
|
2015-04-04 13:13:44 +01:00
|
|
|
(('fsub', a, b), ('fadd', a, ('fneg', b)), 'options->lower_sub'),
|
|
|
|
(('isub', a, b), ('iadd', a, ('ineg', b)), 'options->lower_sub'),
|
2015-01-28 18:52:53 +00:00
|
|
|
(('fneg', a), ('fsub', 0.0, a), 'options->lower_negate'),
|
|
|
|
(('ineg', a), ('isub', 0, a), 'options->lower_negate'),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fadd', a, ('fsub', 0.0, b)), ('fsub', a, b)),
|
2015-01-28 18:52:53 +00:00
|
|
|
(('iadd', a, ('isub', 0, b)), ('isub', a, b)),
|
|
|
|
(('fabs', ('fsub', 0.0, a)), ('fabs', a)),
|
|
|
|
(('iabs', ('isub', 0, a)), ('iabs', a)),
|
2016-01-06 23:30:38 +00:00
|
|
|
|
2016-02-22 23:16:49 +00:00
|
|
|
# Propagate negation up multiplication chains
|
|
|
|
(('fmul', ('fneg', a), b), ('fneg', ('fmul', a, b))),
|
2016-03-31 03:18:16 +01:00
|
|
|
(('imul', ('ineg', a), b), ('ineg', ('imul', a, b))),
|
2016-02-22 23:16:49 +00:00
|
|
|
|
2016-01-06 23:30:38 +00:00
|
|
|
# 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'),
|
2016-01-13 19:08:37 +00:00
|
|
|
|
|
|
|
(('bitfield_insert', 'base', 'insert', 'offset', 'bits'),
|
|
|
|
('bcsel', ('ilt', 31, 'bits'), 'insert',
|
|
|
|
('bfi', ('bfm', 'bits', 'offset'), 'insert', 'base')),
|
|
|
|
'options->lower_bitfield_insert'),
|
2016-01-13 19:09:11 +00:00
|
|
|
|
|
|
|
(('ibitfield_extract', 'value', 'offset', 'bits'),
|
|
|
|
('bcsel', ('ilt', 31, 'bits'), 'value',
|
|
|
|
('ibfe', 'value', 'offset', 'bits')),
|
|
|
|
'options->lower_bitfield_extract'),
|
|
|
|
|
|
|
|
(('ubitfield_extract', 'value', 'offset', 'bits'),
|
|
|
|
('bcsel', ('ult', 31, 'bits'), 'value',
|
|
|
|
('ubfe', 'value', 'offset', 'bits')),
|
|
|
|
'options->lower_bitfield_extract'),
|
2016-01-21 17:09:29 +00:00
|
|
|
|
|
|
|
(('extract_i8', a, b),
|
|
|
|
('ishr', ('ishl', a, ('imul', ('isub', 3, b), 8)), 24),
|
|
|
|
'options->lower_extract_byte'),
|
|
|
|
|
|
|
|
(('extract_u8', a, b),
|
|
|
|
('iand', ('ushr', a, ('imul', b, 8)), 0xff),
|
|
|
|
'options->lower_extract_byte'),
|
|
|
|
|
|
|
|
(('extract_i16', a, b),
|
|
|
|
('ishr', ('ishl', a, ('imul', ('isub', 1, b), 16)), 16),
|
|
|
|
'options->lower_extract_word'),
|
|
|
|
|
|
|
|
(('extract_u16', a, b),
|
|
|
|
('iand', ('ushr', a, ('imul', b, 16)), 0xffff),
|
|
|
|
'options->lower_extract_word'),
|
2016-01-25 19:05:52 +00:00
|
|
|
|
|
|
|
(('pack_unorm_2x16', 'v'),
|
|
|
|
('pack_uvec2_to_uint',
|
|
|
|
('f2u', ('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)))),
|
|
|
|
'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)))),
|
|
|
|
'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)))),
|
|
|
|
'options->lower_pack_snorm_4x8'),
|
2016-01-25 19:07:02 +00:00
|
|
|
|
|
|
|
(('unpack_unorm_2x16', 'v'),
|
|
|
|
('fdiv', ('u2f', ('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))),
|
|
|
|
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))),
|
|
|
|
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))),
|
|
|
|
127.0))),
|
|
|
|
'options->lower_unpack_snorm_4x8'),
|
2014-12-12 19:13:10 +00:00
|
|
|
]
|
|
|
|
|
2016-02-02 00:44:18 +00:00
|
|
|
# Unreal Engine 4 demo applications open-codes bitfieldReverse()
|
|
|
|
def bitfield_reverse(u):
|
|
|
|
step1 = ('ior', ('ishl', u, 16), ('ushr', u, 16))
|
|
|
|
step2 = ('ior', ('ishl', ('iand', step1, 0x00ff00ff), 8), ('ushr', ('iand', step1, 0xff00ff00), 8))
|
|
|
|
step3 = ('ior', ('ishl', ('iand', step2, 0x0f0f0f0f), 4), ('ushr', ('iand', step2, 0xf0f0f0f0), 4))
|
|
|
|
step4 = ('ior', ('ishl', ('iand', step3, 0x33333333), 2), ('ushr', ('iand', step3, 0xcccccccc), 2))
|
|
|
|
step5 = ('ior', ('ishl', ('iand', step4, 0x55555555), 1), ('ushr', ('iand', step4, 0xaaaaaaaa), 1))
|
|
|
|
|
|
|
|
return step5
|
|
|
|
|
|
|
|
optimizations += [(bitfield_reverse('x'), ('bitfield_reverse', 'x'))]
|
|
|
|
|
|
|
|
|
2015-01-29 00:53:51 +00:00
|
|
|
# Add optimizations to handle the case where the result of a ternary is
|
|
|
|
# compared to a constant. This way we can take things like
|
|
|
|
#
|
|
|
|
# (a ? 0 : 1) > 0
|
|
|
|
#
|
|
|
|
# and turn it into
|
|
|
|
#
|
|
|
|
# a ? (0 > 0) : (1 > 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',
|
|
|
|
'ilt', 'ige', 'ieq', 'ine', 'ult', 'uge']:
|
|
|
|
optimizations += [
|
|
|
|
((op, ('bcsel', 'a', '#b', '#c'), '#d'),
|
|
|
|
('bcsel', 'a', (op, 'b', 'd'), (op, 'c', 'd'))),
|
|
|
|
((op, '#d', ('bcsel', a, '#b', '#c')),
|
|
|
|
('bcsel', 'a', (op, 'd', 'b'), (op, 'd', 'c'))),
|
|
|
|
]
|
|
|
|
|
2015-03-24 00:11:49 +00:00
|
|
|
# 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
|
|
|
|
# more easily optimizable.
|
|
|
|
late_optimizations = [
|
2016-03-17 18:31:48 +00:00
|
|
|
# Most of these optimizations aren't quite safe when you get infinity or
|
|
|
|
# Nan involved but the first one should be fine.
|
2015-03-24 00:36:22 +00:00
|
|
|
(('flt', ('fadd', a, b), 0.0), ('flt', a, ('fneg', b))),
|
2016-03-17 18:31:48 +00:00
|
|
|
(('~fge', ('fadd', a, b), 0.0), ('fge', a, ('fneg', b))),
|
|
|
|
(('~feq', ('fadd', a, b), 0.0), ('feq', a, ('fneg', b))),
|
|
|
|
(('~fne', ('fadd', a, b), 0.0), ('fne', a, ('fneg', b))),
|
|
|
|
|
2015-09-10 18:51:46 +01:00
|
|
|
(('fdot2', a, b), ('fdot_replicated2', a, b), 'options->fdot_replicates'),
|
|
|
|
(('fdot3', a, b), ('fdot_replicated3', a, b), 'options->fdot_replicates'),
|
|
|
|
(('fdot4', a, b), ('fdot_replicated4', a, b), 'options->fdot_replicates'),
|
2015-09-23 00:54:27 +01:00
|
|
|
(('fdph', a, b), ('fdph_replicated', a, b), 'options->fdot_replicates'),
|
2015-03-24 00:11:49 +00:00
|
|
|
]
|
|
|
|
|
2014-12-12 19:13:10 +00:00
|
|
|
print nir_algebraic.AlgebraicPass("nir_opt_algebraic", optimizations).render()
|
2015-03-24 00:11:49 +00:00
|
|
|
print nir_algebraic.AlgebraicPass("nir_opt_algebraic_late",
|
|
|
|
late_optimizations).render()
|