nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
/*
|
|
|
|
|
* Copyright © 2019 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.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#include "nir/nir_builder.h"
|
2023-08-08 12:00:35 -05:00
|
|
|
#include "nir.h"
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
#include "nir_constant_expressions.h"
|
|
|
|
|
#include "nir_control_flow.h"
|
|
|
|
|
#include "nir_loop_analyze.h"
|
|
|
|
|
|
|
|
|
|
static bool
|
|
|
|
|
is_two_src_comparison(const nir_alu_instr *instr)
|
|
|
|
|
{
|
|
|
|
|
switch (instr->op) {
|
|
|
|
|
case nir_op_flt:
|
|
|
|
|
case nir_op_flt32:
|
|
|
|
|
case nir_op_fge:
|
|
|
|
|
case nir_op_fge32:
|
|
|
|
|
case nir_op_feq:
|
|
|
|
|
case nir_op_feq32:
|
2020-08-18 19:51:57 +02:00
|
|
|
case nir_op_fneu:
|
|
|
|
|
case nir_op_fneu32:
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
case nir_op_ilt:
|
|
|
|
|
case nir_op_ilt32:
|
|
|
|
|
case nir_op_ult:
|
|
|
|
|
case nir_op_ult32:
|
|
|
|
|
case nir_op_ige:
|
|
|
|
|
case nir_op_ige32:
|
|
|
|
|
case nir_op_uge:
|
|
|
|
|
case nir_op_uge32:
|
|
|
|
|
case nir_op_ieq:
|
|
|
|
|
case nir_op_ieq32:
|
|
|
|
|
case nir_op_ine:
|
|
|
|
|
case nir_op_ine32:
|
|
|
|
|
return true;
|
|
|
|
|
default:
|
|
|
|
|
return false;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
nir/rematerialize: Rematerialize ALUs used only by compares with zero
This was 4th on the list of things to try in 3ee2e84c608 ("nir:
Rematerialize compare instructions"). This is implemented as a separate
subpass that tries to find ALU instructions (with restrictions) that are
only used by comparisons with zero that are in turn only used as
conditions for bcsel or if-statements.
There are two restrictions implemented. One of the sources must be a
constant. This is done in an attempt to prevent increasing register
pressure. Additionally, the opcode of the instruction must be one that
has a high probablility of getting a conditional modifier on Intel
GPUs. Not all instructions can have a conditional modifiers (e.g., min
and max), so I don't think there is any benefit to moving these
instructions.
v2: Rebase on many, many recent NIR infrastructure changes.
v3: Make data in commit message more clear. Suggested by Matt. Rebase on
b5d6b7c402a ("nir: Drop most uses if nir_instr_rewrite_src()").
All of the affected shaders on ILK and G45 are in CS:GO. There is some
brief analysis of the changes in the MR.
Reviewed-by: Matt Tuner <mattst88@gmail.com>
Shader-db results:
DG2
total instructions in shared programs: 22824637 -> 22824258 (<.01%)
instructions in affected programs: 365742 -> 365363 (-0.10%)
helped: 190 / HURT: 97
total cycles in shared programs: 832186193 -> 832157290 (<.01%)
cycles in affected programs: 41245259 -> 41216356 (-0.07%)
helped: 208 / HURT: 117
total spills in shared programs: 4072 -> 4060 (-0.29%)
spills in affected programs: 366 -> 354 (-3.28%)
helped: 4 / HURT: 2
total fills in shared programs: 3601 -> 3607 (0.17%)
fills in affected programs: 708 -> 714 (0.85%)
helped: 4 / HURT: 2
LOST: 0
GAINED: 1
Tiger Lake and Ice Lake had similar results. (Ice Lake shown)
total instructions in shared programs: 20320934 -> 20320689 (<.01%)
instructions in affected programs: 236592 -> 236347 (-0.10%)
helped: 176 / HURT: 29
total cycles in shared programs: 849846341 -> 849843856 (<.01%)
cycles in affected programs: 41277336 -> 41274851 (<.01%)
helped: 195 / HURT: 110
LOST: 0
GAINED: 1
Skylake
total instructions in shared programs: 18550811 -> 18550470 (<.01%)
instructions in affected programs: 233908 -> 233567 (-0.15%)
helped: 182 / HURT: 25
total cycles in shared programs: 835910983 -> 835889167 (<.01%)
cycles in affected programs: 38764359 -> 38742543 (-0.06%)
helped: 207/ HURT: 94
total spills in shared programs: 4522 -> 4506 (-0.35%)
spills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
total fills in shared programs: 5296 -> 5280 (-0.30%)
fills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
LOST: 0
GAINED: 1
Broadwell
total instructions in shared programs: 18199130 -> 18197920 (<.01%)
instructions in affected programs: 214664 -> 213454 (-0.56%)
helped: 191 / HURT: 0
total cycles in shared programs: 935131908 -> 934870248 (-0.03%)
cycles in affected programs: 75770568 -> 75508908 (-0.35%)
helped: 203 / HURT: 84
total spills in shared programs: 13896 -> 13734 (-1.17%)
spills in affected programs: 162 -> 0
helped: 3 / HURT: 0
total fills in shared programs: 16989 -> 16761 (-1.34%)
fills in affected programs: 228 -> 0
helped: 3 / HURT: 0
Haswell
total instructions in shared programs: 16969502 -> 16969085 (<.01%)
instructions in affected programs: 185498 -> 185081 (-0.22%)
helped: 121 / HURT: 1
total cycles in shared programs: 925290863 -> 924806827 (-0.05%)
cycles in affected programs: 30200863 -> 29716827 (-1.60%)
helped: 100 / HURT: 85
total spills in shared programs: 13565 -> 13533 (-0.24%)
spills in affected programs: 736 -> 704 (-4.35%)
helped: 8 / HURT: 0
total fills in shared programs: 15468 -> 15436 (-0.21%)
fills in affected programs: 740 -> 708 (-4.32%)
helped: 8 / HURT: 0
LOST: 0
GAINED: 1
Ivy Bridge
total instructions in shared programs: 15839127 -> 15838947 (<.01%)
instructions in affected programs: 77776 -> 77596 (-0.23%)
helped: 58 / HURT: 0
total cycles in shared programs: 459852774 -> 459739770 (-0.02%)
cycles in affected programs: 11970210 -> 11857206 (-0.94%)
helped: 79 / HURT: 53
Sandy Bridge
total instructions in shared programs: 14106847 -> 14106831 (<.01%)
instructions in affected programs: 1611 -> 1595 (-0.99%)
helped: 10 / HURT: 0
total cycles in shared programs: 775004024 -> 775007516 (<.01%)
cycles in affected programs: 2530686 -> 2534178 (0.14%)
helped: 55 / HURT: 48
Iron Lake
total cycles in shared programs: 257753356 -> 257754900 (<.01%)
cycles in affected programs: 2977374 -> 2978918 (0.05%)
helped: 12 / HURT: 106
GM45
total cycles in shared programs: 169711382 -> 169712816 (<.01%)
cycles in affected programs: 2402070 -> 2403504 (0.06%)
helped: 12 / HURT: 57
Fossil-db results:
All Intel platforms had similar results. (DG2 shown)
Totals:
Instrs: 193884596 -> 193465896 (-0.22%); split: -0.25%, +0.03%
Cycles: 14050193354 -> 14048194826 (-0.01%); split: -0.34%, +0.33%
Spill count: 114944 -> 100449 (-12.61%); split: -13.59%, +0.98%
Fill count: 201525 -> 179534 (-10.91%); split: -11.22%, +0.31%
Scratch Memory Size: 10028032 -> 8468480 (-15.55%)
Totals from 16912 (2.59% of 653124) affected shaders:
Instrs: 34173709 -> 33755009 (-1.23%); split: -1.41%, +0.19%
Cycles: 2945969110 -> 2943970582 (-0.07%); split: -1.62%, +1.55%
Spill count: 97753 -> 83258 (-14.83%); split: -15.98%, +1.15%
Fill count: 176355 -> 154364 (-12.47%); split: -12.82%, +0.35%
Scratch Memory Size: 8619008 -> 7059456 (-18.09%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20176>
2022-12-05 14:18:33 -08:00
|
|
|
static inline bool
|
|
|
|
|
is_zero(const nir_alu_instr *instr, unsigned src, unsigned num_components,
|
|
|
|
|
const uint8_t *swizzle)
|
|
|
|
|
{
|
|
|
|
|
/* only constant srcs: */
|
|
|
|
|
if (!nir_src_is_const(instr->src[src].src))
|
|
|
|
|
return false;
|
|
|
|
|
|
|
|
|
|
for (unsigned i = 0; i < num_components; i++) {
|
|
|
|
|
nir_alu_type type = nir_op_infos[instr->op].input_types[src];
|
|
|
|
|
switch (nir_alu_type_get_base_type(type)) {
|
|
|
|
|
case nir_type_int:
|
|
|
|
|
case nir_type_uint: {
|
|
|
|
|
if (nir_src_comp_as_int(instr->src[src].src, swizzle[i]) != 0)
|
|
|
|
|
return false;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
case nir_type_float: {
|
|
|
|
|
if (nir_src_comp_as_float(instr->src[src].src, swizzle[i]) != 0)
|
|
|
|
|
return false;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
default:
|
|
|
|
|
return false;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return true;
|
|
|
|
|
}
|
|
|
|
|
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
static bool
|
|
|
|
|
all_uses_are_bcsel(const nir_alu_instr *instr)
|
|
|
|
|
{
|
2023-08-14 11:43:35 -05:00
|
|
|
nir_foreach_use(use, &instr->def) {
|
2023-08-14 09:58:47 -04:00
|
|
|
if (nir_src_parent_instr(use)->type != nir_instr_type_alu)
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
return false;
|
|
|
|
|
|
2023-08-14 09:58:47 -04:00
|
|
|
nir_alu_instr *const alu = nir_instr_as_alu(nir_src_parent_instr(use));
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
if (alu->op != nir_op_bcsel &&
|
|
|
|
|
alu->op != nir_op_b32csel)
|
|
|
|
|
return false;
|
|
|
|
|
|
|
|
|
|
/* Not only must the result be used by a bcsel, but it must be used as
|
|
|
|
|
* the first source (the condition).
|
|
|
|
|
*/
|
2023-08-14 11:43:35 -05:00
|
|
|
if (alu->src[0].src.ssa != &instr->def)
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
return false;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return true;
|
|
|
|
|
}
|
|
|
|
|
|
nir/rematerialize: Rematerialize ALUs used only by compares with zero
This was 4th on the list of things to try in 3ee2e84c608 ("nir:
Rematerialize compare instructions"). This is implemented as a separate
subpass that tries to find ALU instructions (with restrictions) that are
only used by comparisons with zero that are in turn only used as
conditions for bcsel or if-statements.
There are two restrictions implemented. One of the sources must be a
constant. This is done in an attempt to prevent increasing register
pressure. Additionally, the opcode of the instruction must be one that
has a high probablility of getting a conditional modifier on Intel
GPUs. Not all instructions can have a conditional modifiers (e.g., min
and max), so I don't think there is any benefit to moving these
instructions.
v2: Rebase on many, many recent NIR infrastructure changes.
v3: Make data in commit message more clear. Suggested by Matt. Rebase on
b5d6b7c402a ("nir: Drop most uses if nir_instr_rewrite_src()").
All of the affected shaders on ILK and G45 are in CS:GO. There is some
brief analysis of the changes in the MR.
Reviewed-by: Matt Tuner <mattst88@gmail.com>
Shader-db results:
DG2
total instructions in shared programs: 22824637 -> 22824258 (<.01%)
instructions in affected programs: 365742 -> 365363 (-0.10%)
helped: 190 / HURT: 97
total cycles in shared programs: 832186193 -> 832157290 (<.01%)
cycles in affected programs: 41245259 -> 41216356 (-0.07%)
helped: 208 / HURT: 117
total spills in shared programs: 4072 -> 4060 (-0.29%)
spills in affected programs: 366 -> 354 (-3.28%)
helped: 4 / HURT: 2
total fills in shared programs: 3601 -> 3607 (0.17%)
fills in affected programs: 708 -> 714 (0.85%)
helped: 4 / HURT: 2
LOST: 0
GAINED: 1
Tiger Lake and Ice Lake had similar results. (Ice Lake shown)
total instructions in shared programs: 20320934 -> 20320689 (<.01%)
instructions in affected programs: 236592 -> 236347 (-0.10%)
helped: 176 / HURT: 29
total cycles in shared programs: 849846341 -> 849843856 (<.01%)
cycles in affected programs: 41277336 -> 41274851 (<.01%)
helped: 195 / HURT: 110
LOST: 0
GAINED: 1
Skylake
total instructions in shared programs: 18550811 -> 18550470 (<.01%)
instructions in affected programs: 233908 -> 233567 (-0.15%)
helped: 182 / HURT: 25
total cycles in shared programs: 835910983 -> 835889167 (<.01%)
cycles in affected programs: 38764359 -> 38742543 (-0.06%)
helped: 207/ HURT: 94
total spills in shared programs: 4522 -> 4506 (-0.35%)
spills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
total fills in shared programs: 5296 -> 5280 (-0.30%)
fills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
LOST: 0
GAINED: 1
Broadwell
total instructions in shared programs: 18199130 -> 18197920 (<.01%)
instructions in affected programs: 214664 -> 213454 (-0.56%)
helped: 191 / HURT: 0
total cycles in shared programs: 935131908 -> 934870248 (-0.03%)
cycles in affected programs: 75770568 -> 75508908 (-0.35%)
helped: 203 / HURT: 84
total spills in shared programs: 13896 -> 13734 (-1.17%)
spills in affected programs: 162 -> 0
helped: 3 / HURT: 0
total fills in shared programs: 16989 -> 16761 (-1.34%)
fills in affected programs: 228 -> 0
helped: 3 / HURT: 0
Haswell
total instructions in shared programs: 16969502 -> 16969085 (<.01%)
instructions in affected programs: 185498 -> 185081 (-0.22%)
helped: 121 / HURT: 1
total cycles in shared programs: 925290863 -> 924806827 (-0.05%)
cycles in affected programs: 30200863 -> 29716827 (-1.60%)
helped: 100 / HURT: 85
total spills in shared programs: 13565 -> 13533 (-0.24%)
spills in affected programs: 736 -> 704 (-4.35%)
helped: 8 / HURT: 0
total fills in shared programs: 15468 -> 15436 (-0.21%)
fills in affected programs: 740 -> 708 (-4.32%)
helped: 8 / HURT: 0
LOST: 0
GAINED: 1
Ivy Bridge
total instructions in shared programs: 15839127 -> 15838947 (<.01%)
instructions in affected programs: 77776 -> 77596 (-0.23%)
helped: 58 / HURT: 0
total cycles in shared programs: 459852774 -> 459739770 (-0.02%)
cycles in affected programs: 11970210 -> 11857206 (-0.94%)
helped: 79 / HURT: 53
Sandy Bridge
total instructions in shared programs: 14106847 -> 14106831 (<.01%)
instructions in affected programs: 1611 -> 1595 (-0.99%)
helped: 10 / HURT: 0
total cycles in shared programs: 775004024 -> 775007516 (<.01%)
cycles in affected programs: 2530686 -> 2534178 (0.14%)
helped: 55 / HURT: 48
Iron Lake
total cycles in shared programs: 257753356 -> 257754900 (<.01%)
cycles in affected programs: 2977374 -> 2978918 (0.05%)
helped: 12 / HURT: 106
GM45
total cycles in shared programs: 169711382 -> 169712816 (<.01%)
cycles in affected programs: 2402070 -> 2403504 (0.06%)
helped: 12 / HURT: 57
Fossil-db results:
All Intel platforms had similar results. (DG2 shown)
Totals:
Instrs: 193884596 -> 193465896 (-0.22%); split: -0.25%, +0.03%
Cycles: 14050193354 -> 14048194826 (-0.01%); split: -0.34%, +0.33%
Spill count: 114944 -> 100449 (-12.61%); split: -13.59%, +0.98%
Fill count: 201525 -> 179534 (-10.91%); split: -11.22%, +0.31%
Scratch Memory Size: 10028032 -> 8468480 (-15.55%)
Totals from 16912 (2.59% of 653124) affected shaders:
Instrs: 34173709 -> 33755009 (-1.23%); split: -1.41%, +0.19%
Cycles: 2945969110 -> 2943970582 (-0.07%); split: -1.62%, +1.55%
Spill count: 97753 -> 83258 (-14.83%); split: -15.98%, +1.15%
Fill count: 176355 -> 154364 (-12.47%); split: -12.82%, +0.35%
Scratch Memory Size: 8619008 -> 7059456 (-18.09%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20176>
2022-12-05 14:18:33 -08:00
|
|
|
static bool
|
|
|
|
|
all_uses_are_compare_with_zero(const nir_alu_instr *instr)
|
|
|
|
|
{
|
|
|
|
|
nir_foreach_use(use, &instr->def) {
|
2023-08-14 09:58:47 -04:00
|
|
|
if (nir_src_parent_instr(use)->type != nir_instr_type_alu)
|
nir/rematerialize: Rematerialize ALUs used only by compares with zero
This was 4th on the list of things to try in 3ee2e84c608 ("nir:
Rematerialize compare instructions"). This is implemented as a separate
subpass that tries to find ALU instructions (with restrictions) that are
only used by comparisons with zero that are in turn only used as
conditions for bcsel or if-statements.
There are two restrictions implemented. One of the sources must be a
constant. This is done in an attempt to prevent increasing register
pressure. Additionally, the opcode of the instruction must be one that
has a high probablility of getting a conditional modifier on Intel
GPUs. Not all instructions can have a conditional modifiers (e.g., min
and max), so I don't think there is any benefit to moving these
instructions.
v2: Rebase on many, many recent NIR infrastructure changes.
v3: Make data in commit message more clear. Suggested by Matt. Rebase on
b5d6b7c402a ("nir: Drop most uses if nir_instr_rewrite_src()").
All of the affected shaders on ILK and G45 are in CS:GO. There is some
brief analysis of the changes in the MR.
Reviewed-by: Matt Tuner <mattst88@gmail.com>
Shader-db results:
DG2
total instructions in shared programs: 22824637 -> 22824258 (<.01%)
instructions in affected programs: 365742 -> 365363 (-0.10%)
helped: 190 / HURT: 97
total cycles in shared programs: 832186193 -> 832157290 (<.01%)
cycles in affected programs: 41245259 -> 41216356 (-0.07%)
helped: 208 / HURT: 117
total spills in shared programs: 4072 -> 4060 (-0.29%)
spills in affected programs: 366 -> 354 (-3.28%)
helped: 4 / HURT: 2
total fills in shared programs: 3601 -> 3607 (0.17%)
fills in affected programs: 708 -> 714 (0.85%)
helped: 4 / HURT: 2
LOST: 0
GAINED: 1
Tiger Lake and Ice Lake had similar results. (Ice Lake shown)
total instructions in shared programs: 20320934 -> 20320689 (<.01%)
instructions in affected programs: 236592 -> 236347 (-0.10%)
helped: 176 / HURT: 29
total cycles in shared programs: 849846341 -> 849843856 (<.01%)
cycles in affected programs: 41277336 -> 41274851 (<.01%)
helped: 195 / HURT: 110
LOST: 0
GAINED: 1
Skylake
total instructions in shared programs: 18550811 -> 18550470 (<.01%)
instructions in affected programs: 233908 -> 233567 (-0.15%)
helped: 182 / HURT: 25
total cycles in shared programs: 835910983 -> 835889167 (<.01%)
cycles in affected programs: 38764359 -> 38742543 (-0.06%)
helped: 207/ HURT: 94
total spills in shared programs: 4522 -> 4506 (-0.35%)
spills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
total fills in shared programs: 5296 -> 5280 (-0.30%)
fills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
LOST: 0
GAINED: 1
Broadwell
total instructions in shared programs: 18199130 -> 18197920 (<.01%)
instructions in affected programs: 214664 -> 213454 (-0.56%)
helped: 191 / HURT: 0
total cycles in shared programs: 935131908 -> 934870248 (-0.03%)
cycles in affected programs: 75770568 -> 75508908 (-0.35%)
helped: 203 / HURT: 84
total spills in shared programs: 13896 -> 13734 (-1.17%)
spills in affected programs: 162 -> 0
helped: 3 / HURT: 0
total fills in shared programs: 16989 -> 16761 (-1.34%)
fills in affected programs: 228 -> 0
helped: 3 / HURT: 0
Haswell
total instructions in shared programs: 16969502 -> 16969085 (<.01%)
instructions in affected programs: 185498 -> 185081 (-0.22%)
helped: 121 / HURT: 1
total cycles in shared programs: 925290863 -> 924806827 (-0.05%)
cycles in affected programs: 30200863 -> 29716827 (-1.60%)
helped: 100 / HURT: 85
total spills in shared programs: 13565 -> 13533 (-0.24%)
spills in affected programs: 736 -> 704 (-4.35%)
helped: 8 / HURT: 0
total fills in shared programs: 15468 -> 15436 (-0.21%)
fills in affected programs: 740 -> 708 (-4.32%)
helped: 8 / HURT: 0
LOST: 0
GAINED: 1
Ivy Bridge
total instructions in shared programs: 15839127 -> 15838947 (<.01%)
instructions in affected programs: 77776 -> 77596 (-0.23%)
helped: 58 / HURT: 0
total cycles in shared programs: 459852774 -> 459739770 (-0.02%)
cycles in affected programs: 11970210 -> 11857206 (-0.94%)
helped: 79 / HURT: 53
Sandy Bridge
total instructions in shared programs: 14106847 -> 14106831 (<.01%)
instructions in affected programs: 1611 -> 1595 (-0.99%)
helped: 10 / HURT: 0
total cycles in shared programs: 775004024 -> 775007516 (<.01%)
cycles in affected programs: 2530686 -> 2534178 (0.14%)
helped: 55 / HURT: 48
Iron Lake
total cycles in shared programs: 257753356 -> 257754900 (<.01%)
cycles in affected programs: 2977374 -> 2978918 (0.05%)
helped: 12 / HURT: 106
GM45
total cycles in shared programs: 169711382 -> 169712816 (<.01%)
cycles in affected programs: 2402070 -> 2403504 (0.06%)
helped: 12 / HURT: 57
Fossil-db results:
All Intel platforms had similar results. (DG2 shown)
Totals:
Instrs: 193884596 -> 193465896 (-0.22%); split: -0.25%, +0.03%
Cycles: 14050193354 -> 14048194826 (-0.01%); split: -0.34%, +0.33%
Spill count: 114944 -> 100449 (-12.61%); split: -13.59%, +0.98%
Fill count: 201525 -> 179534 (-10.91%); split: -11.22%, +0.31%
Scratch Memory Size: 10028032 -> 8468480 (-15.55%)
Totals from 16912 (2.59% of 653124) affected shaders:
Instrs: 34173709 -> 33755009 (-1.23%); split: -1.41%, +0.19%
Cycles: 2945969110 -> 2943970582 (-0.07%); split: -1.62%, +1.55%
Spill count: 97753 -> 83258 (-14.83%); split: -15.98%, +1.15%
Fill count: 176355 -> 154364 (-12.47%); split: -12.82%, +0.35%
Scratch Memory Size: 8619008 -> 7059456 (-18.09%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20176>
2022-12-05 14:18:33 -08:00
|
|
|
return false;
|
|
|
|
|
|
2023-08-14 09:58:47 -04:00
|
|
|
nir_alu_instr *const alu = nir_instr_as_alu(nir_src_parent_instr(use));
|
nir/rematerialize: Rematerialize ALUs used only by compares with zero
This was 4th on the list of things to try in 3ee2e84c608 ("nir:
Rematerialize compare instructions"). This is implemented as a separate
subpass that tries to find ALU instructions (with restrictions) that are
only used by comparisons with zero that are in turn only used as
conditions for bcsel or if-statements.
There are two restrictions implemented. One of the sources must be a
constant. This is done in an attempt to prevent increasing register
pressure. Additionally, the opcode of the instruction must be one that
has a high probablility of getting a conditional modifier on Intel
GPUs. Not all instructions can have a conditional modifiers (e.g., min
and max), so I don't think there is any benefit to moving these
instructions.
v2: Rebase on many, many recent NIR infrastructure changes.
v3: Make data in commit message more clear. Suggested by Matt. Rebase on
b5d6b7c402a ("nir: Drop most uses if nir_instr_rewrite_src()").
All of the affected shaders on ILK and G45 are in CS:GO. There is some
brief analysis of the changes in the MR.
Reviewed-by: Matt Tuner <mattst88@gmail.com>
Shader-db results:
DG2
total instructions in shared programs: 22824637 -> 22824258 (<.01%)
instructions in affected programs: 365742 -> 365363 (-0.10%)
helped: 190 / HURT: 97
total cycles in shared programs: 832186193 -> 832157290 (<.01%)
cycles in affected programs: 41245259 -> 41216356 (-0.07%)
helped: 208 / HURT: 117
total spills in shared programs: 4072 -> 4060 (-0.29%)
spills in affected programs: 366 -> 354 (-3.28%)
helped: 4 / HURT: 2
total fills in shared programs: 3601 -> 3607 (0.17%)
fills in affected programs: 708 -> 714 (0.85%)
helped: 4 / HURT: 2
LOST: 0
GAINED: 1
Tiger Lake and Ice Lake had similar results. (Ice Lake shown)
total instructions in shared programs: 20320934 -> 20320689 (<.01%)
instructions in affected programs: 236592 -> 236347 (-0.10%)
helped: 176 / HURT: 29
total cycles in shared programs: 849846341 -> 849843856 (<.01%)
cycles in affected programs: 41277336 -> 41274851 (<.01%)
helped: 195 / HURT: 110
LOST: 0
GAINED: 1
Skylake
total instructions in shared programs: 18550811 -> 18550470 (<.01%)
instructions in affected programs: 233908 -> 233567 (-0.15%)
helped: 182 / HURT: 25
total cycles in shared programs: 835910983 -> 835889167 (<.01%)
cycles in affected programs: 38764359 -> 38742543 (-0.06%)
helped: 207/ HURT: 94
total spills in shared programs: 4522 -> 4506 (-0.35%)
spills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
total fills in shared programs: 5296 -> 5280 (-0.30%)
fills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
LOST: 0
GAINED: 1
Broadwell
total instructions in shared programs: 18199130 -> 18197920 (<.01%)
instructions in affected programs: 214664 -> 213454 (-0.56%)
helped: 191 / HURT: 0
total cycles in shared programs: 935131908 -> 934870248 (-0.03%)
cycles in affected programs: 75770568 -> 75508908 (-0.35%)
helped: 203 / HURT: 84
total spills in shared programs: 13896 -> 13734 (-1.17%)
spills in affected programs: 162 -> 0
helped: 3 / HURT: 0
total fills in shared programs: 16989 -> 16761 (-1.34%)
fills in affected programs: 228 -> 0
helped: 3 / HURT: 0
Haswell
total instructions in shared programs: 16969502 -> 16969085 (<.01%)
instructions in affected programs: 185498 -> 185081 (-0.22%)
helped: 121 / HURT: 1
total cycles in shared programs: 925290863 -> 924806827 (-0.05%)
cycles in affected programs: 30200863 -> 29716827 (-1.60%)
helped: 100 / HURT: 85
total spills in shared programs: 13565 -> 13533 (-0.24%)
spills in affected programs: 736 -> 704 (-4.35%)
helped: 8 / HURT: 0
total fills in shared programs: 15468 -> 15436 (-0.21%)
fills in affected programs: 740 -> 708 (-4.32%)
helped: 8 / HURT: 0
LOST: 0
GAINED: 1
Ivy Bridge
total instructions in shared programs: 15839127 -> 15838947 (<.01%)
instructions in affected programs: 77776 -> 77596 (-0.23%)
helped: 58 / HURT: 0
total cycles in shared programs: 459852774 -> 459739770 (-0.02%)
cycles in affected programs: 11970210 -> 11857206 (-0.94%)
helped: 79 / HURT: 53
Sandy Bridge
total instructions in shared programs: 14106847 -> 14106831 (<.01%)
instructions in affected programs: 1611 -> 1595 (-0.99%)
helped: 10 / HURT: 0
total cycles in shared programs: 775004024 -> 775007516 (<.01%)
cycles in affected programs: 2530686 -> 2534178 (0.14%)
helped: 55 / HURT: 48
Iron Lake
total cycles in shared programs: 257753356 -> 257754900 (<.01%)
cycles in affected programs: 2977374 -> 2978918 (0.05%)
helped: 12 / HURT: 106
GM45
total cycles in shared programs: 169711382 -> 169712816 (<.01%)
cycles in affected programs: 2402070 -> 2403504 (0.06%)
helped: 12 / HURT: 57
Fossil-db results:
All Intel platforms had similar results. (DG2 shown)
Totals:
Instrs: 193884596 -> 193465896 (-0.22%); split: -0.25%, +0.03%
Cycles: 14050193354 -> 14048194826 (-0.01%); split: -0.34%, +0.33%
Spill count: 114944 -> 100449 (-12.61%); split: -13.59%, +0.98%
Fill count: 201525 -> 179534 (-10.91%); split: -11.22%, +0.31%
Scratch Memory Size: 10028032 -> 8468480 (-15.55%)
Totals from 16912 (2.59% of 653124) affected shaders:
Instrs: 34173709 -> 33755009 (-1.23%); split: -1.41%, +0.19%
Cycles: 2945969110 -> 2943970582 (-0.07%); split: -1.62%, +1.55%
Spill count: 97753 -> 83258 (-14.83%); split: -15.98%, +1.15%
Fill count: 176355 -> 154364 (-12.47%); split: -12.82%, +0.35%
Scratch Memory Size: 8619008 -> 7059456 (-18.09%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20176>
2022-12-05 14:18:33 -08:00
|
|
|
if (!is_two_src_comparison(alu))
|
|
|
|
|
return false;
|
|
|
|
|
|
|
|
|
|
if (!is_zero(alu, 0, 1, alu->src[0].swizzle) &&
|
|
|
|
|
!is_zero(alu, 1, 1, alu->src[1].swizzle))
|
|
|
|
|
return false;
|
|
|
|
|
|
|
|
|
|
if (!all_uses_are_bcsel(alu))
|
|
|
|
|
return false;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return true;
|
|
|
|
|
}
|
|
|
|
|
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
static bool
|
|
|
|
|
nir_opt_rematerialize_compares_impl(nir_shader *shader, nir_function_impl *impl)
|
|
|
|
|
{
|
|
|
|
|
bool progress = false;
|
|
|
|
|
|
|
|
|
|
nir_foreach_block(block, impl) {
|
|
|
|
|
nir_foreach_instr(instr, block) {
|
|
|
|
|
if (instr->type != nir_instr_type_alu)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
nir_alu_instr *const alu = nir_instr_as_alu(instr);
|
|
|
|
|
if (!is_two_src_comparison(alu))
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
if (!all_uses_are_bcsel(alu))
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
/* At this point it is known that alu is a comparison instruction
|
|
|
|
|
* that is only used by nir_op_bcsel and possibly by if-statements
|
|
|
|
|
* (though the latter has not been explicitly checked).
|
|
|
|
|
*
|
|
|
|
|
* Iterate through each use of the comparison. For every use (or use
|
|
|
|
|
* by an if-statement) that is in a different block, emit a copy of
|
|
|
|
|
* the comparison. Care must be taken here. The original
|
|
|
|
|
* instruction must be duplicated only once in each block because CSE
|
|
|
|
|
* cannot be run after this pass.
|
|
|
|
|
*/
|
2023-08-14 11:43:35 -05:00
|
|
|
nir_foreach_use_including_if_safe(use, &alu->def) {
|
2023-08-14 09:58:47 -04:00
|
|
|
if (nir_src_is_if(use)) {
|
|
|
|
|
nir_if *const if_stmt = nir_src_parent_if(use);
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
|
2023-04-06 13:19:31 -04:00
|
|
|
nir_block *const prev_block =
|
|
|
|
|
nir_cf_node_as_block(nir_cf_node_prev(&if_stmt->cf_node));
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
|
2023-04-06 13:19:31 -04:00
|
|
|
/* If the compare is from the previous block, don't
|
|
|
|
|
* rematerialize.
|
|
|
|
|
*/
|
|
|
|
|
if (prev_block == alu->instr.block)
|
|
|
|
|
continue;
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
|
2023-04-06 13:19:31 -04:00
|
|
|
nir_alu_instr *clone = nir_alu_instr_clone(shader, alu);
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
|
2023-04-06 13:19:31 -04:00
|
|
|
nir_instr_insert_after_block(prev_block, &clone->instr);
|
|
|
|
|
|
2023-08-17 15:41:41 -05:00
|
|
|
nir_src_rewrite(&if_stmt->condition, &clone->def);
|
2023-04-06 13:19:31 -04:00
|
|
|
progress = true;
|
|
|
|
|
} else {
|
2023-08-14 09:58:47 -04:00
|
|
|
nir_instr *const use_instr = nir_src_parent_instr(use);
|
2023-04-06 13:19:31 -04:00
|
|
|
|
|
|
|
|
/* If the use is in the same block as the def, don't
|
|
|
|
|
* rematerialize.
|
|
|
|
|
*/
|
|
|
|
|
if (use_instr->block == alu->instr.block)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
nir_alu_instr *clone = nir_alu_instr_clone(shader, alu);
|
|
|
|
|
|
|
|
|
|
nir_instr_insert_before(use_instr, &clone->instr);
|
|
|
|
|
|
|
|
|
|
nir_alu_instr *const use_alu = nir_instr_as_alu(use_instr);
|
|
|
|
|
for (unsigned i = 0; i < nir_op_infos[use_alu->op].num_inputs; i++) {
|
2023-08-14 11:43:35 -05:00
|
|
|
if (use_alu->src[i].src.ssa == &alu->def) {
|
2023-08-17 16:27:15 -05:00
|
|
|
nir_src_rewrite(&use_alu->src[i].src, &clone->def);
|
2023-04-06 13:19:31 -04:00
|
|
|
progress = true;
|
|
|
|
|
}
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (progress) {
|
2024-06-16 16:32:01 -04:00
|
|
|
nir_metadata_preserve(impl, nir_metadata_control_flow);
|
2020-05-21 20:41:12 -05:00
|
|
|
} else {
|
|
|
|
|
nir_metadata_preserve(impl, nir_metadata_all);
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return progress;
|
|
|
|
|
}
|
|
|
|
|
|
nir/rematerialize: Rematerialize ALUs used only by compares with zero
This was 4th on the list of things to try in 3ee2e84c608 ("nir:
Rematerialize compare instructions"). This is implemented as a separate
subpass that tries to find ALU instructions (with restrictions) that are
only used by comparisons with zero that are in turn only used as
conditions for bcsel or if-statements.
There are two restrictions implemented. One of the sources must be a
constant. This is done in an attempt to prevent increasing register
pressure. Additionally, the opcode of the instruction must be one that
has a high probablility of getting a conditional modifier on Intel
GPUs. Not all instructions can have a conditional modifiers (e.g., min
and max), so I don't think there is any benefit to moving these
instructions.
v2: Rebase on many, many recent NIR infrastructure changes.
v3: Make data in commit message more clear. Suggested by Matt. Rebase on
b5d6b7c402a ("nir: Drop most uses if nir_instr_rewrite_src()").
All of the affected shaders on ILK and G45 are in CS:GO. There is some
brief analysis of the changes in the MR.
Reviewed-by: Matt Tuner <mattst88@gmail.com>
Shader-db results:
DG2
total instructions in shared programs: 22824637 -> 22824258 (<.01%)
instructions in affected programs: 365742 -> 365363 (-0.10%)
helped: 190 / HURT: 97
total cycles in shared programs: 832186193 -> 832157290 (<.01%)
cycles in affected programs: 41245259 -> 41216356 (-0.07%)
helped: 208 / HURT: 117
total spills in shared programs: 4072 -> 4060 (-0.29%)
spills in affected programs: 366 -> 354 (-3.28%)
helped: 4 / HURT: 2
total fills in shared programs: 3601 -> 3607 (0.17%)
fills in affected programs: 708 -> 714 (0.85%)
helped: 4 / HURT: 2
LOST: 0
GAINED: 1
Tiger Lake and Ice Lake had similar results. (Ice Lake shown)
total instructions in shared programs: 20320934 -> 20320689 (<.01%)
instructions in affected programs: 236592 -> 236347 (-0.10%)
helped: 176 / HURT: 29
total cycles in shared programs: 849846341 -> 849843856 (<.01%)
cycles in affected programs: 41277336 -> 41274851 (<.01%)
helped: 195 / HURT: 110
LOST: 0
GAINED: 1
Skylake
total instructions in shared programs: 18550811 -> 18550470 (<.01%)
instructions in affected programs: 233908 -> 233567 (-0.15%)
helped: 182 / HURT: 25
total cycles in shared programs: 835910983 -> 835889167 (<.01%)
cycles in affected programs: 38764359 -> 38742543 (-0.06%)
helped: 207/ HURT: 94
total spills in shared programs: 4522 -> 4506 (-0.35%)
spills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
total fills in shared programs: 5296 -> 5280 (-0.30%)
fills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
LOST: 0
GAINED: 1
Broadwell
total instructions in shared programs: 18199130 -> 18197920 (<.01%)
instructions in affected programs: 214664 -> 213454 (-0.56%)
helped: 191 / HURT: 0
total cycles in shared programs: 935131908 -> 934870248 (-0.03%)
cycles in affected programs: 75770568 -> 75508908 (-0.35%)
helped: 203 / HURT: 84
total spills in shared programs: 13896 -> 13734 (-1.17%)
spills in affected programs: 162 -> 0
helped: 3 / HURT: 0
total fills in shared programs: 16989 -> 16761 (-1.34%)
fills in affected programs: 228 -> 0
helped: 3 / HURT: 0
Haswell
total instructions in shared programs: 16969502 -> 16969085 (<.01%)
instructions in affected programs: 185498 -> 185081 (-0.22%)
helped: 121 / HURT: 1
total cycles in shared programs: 925290863 -> 924806827 (-0.05%)
cycles in affected programs: 30200863 -> 29716827 (-1.60%)
helped: 100 / HURT: 85
total spills in shared programs: 13565 -> 13533 (-0.24%)
spills in affected programs: 736 -> 704 (-4.35%)
helped: 8 / HURT: 0
total fills in shared programs: 15468 -> 15436 (-0.21%)
fills in affected programs: 740 -> 708 (-4.32%)
helped: 8 / HURT: 0
LOST: 0
GAINED: 1
Ivy Bridge
total instructions in shared programs: 15839127 -> 15838947 (<.01%)
instructions in affected programs: 77776 -> 77596 (-0.23%)
helped: 58 / HURT: 0
total cycles in shared programs: 459852774 -> 459739770 (-0.02%)
cycles in affected programs: 11970210 -> 11857206 (-0.94%)
helped: 79 / HURT: 53
Sandy Bridge
total instructions in shared programs: 14106847 -> 14106831 (<.01%)
instructions in affected programs: 1611 -> 1595 (-0.99%)
helped: 10 / HURT: 0
total cycles in shared programs: 775004024 -> 775007516 (<.01%)
cycles in affected programs: 2530686 -> 2534178 (0.14%)
helped: 55 / HURT: 48
Iron Lake
total cycles in shared programs: 257753356 -> 257754900 (<.01%)
cycles in affected programs: 2977374 -> 2978918 (0.05%)
helped: 12 / HURT: 106
GM45
total cycles in shared programs: 169711382 -> 169712816 (<.01%)
cycles in affected programs: 2402070 -> 2403504 (0.06%)
helped: 12 / HURT: 57
Fossil-db results:
All Intel platforms had similar results. (DG2 shown)
Totals:
Instrs: 193884596 -> 193465896 (-0.22%); split: -0.25%, +0.03%
Cycles: 14050193354 -> 14048194826 (-0.01%); split: -0.34%, +0.33%
Spill count: 114944 -> 100449 (-12.61%); split: -13.59%, +0.98%
Fill count: 201525 -> 179534 (-10.91%); split: -11.22%, +0.31%
Scratch Memory Size: 10028032 -> 8468480 (-15.55%)
Totals from 16912 (2.59% of 653124) affected shaders:
Instrs: 34173709 -> 33755009 (-1.23%); split: -1.41%, +0.19%
Cycles: 2945969110 -> 2943970582 (-0.07%); split: -1.62%, +1.55%
Spill count: 97753 -> 83258 (-14.83%); split: -15.98%, +1.15%
Fill count: 176355 -> 154364 (-12.47%); split: -12.82%, +0.35%
Scratch Memory Size: 8619008 -> 7059456 (-18.09%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20176>
2022-12-05 14:18:33 -08:00
|
|
|
static bool
|
|
|
|
|
nir_opt_rematerialize_alu_impl(nir_shader *shader, nir_function_impl *impl)
|
|
|
|
|
{
|
|
|
|
|
bool progress = false;
|
|
|
|
|
|
|
|
|
|
nir_foreach_block(block, impl) {
|
|
|
|
|
nir_foreach_instr(instr, block) {
|
|
|
|
|
if (instr->type != nir_instr_type_alu)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
nir_alu_instr *const alu = nir_instr_as_alu(instr);
|
|
|
|
|
|
|
|
|
|
/* This list only include ALU ops that are likely to be able to have
|
|
|
|
|
* cmod propagation on Intel GPUs.
|
|
|
|
|
*/
|
|
|
|
|
switch (alu->op) {
|
|
|
|
|
case nir_op_ineg:
|
|
|
|
|
case nir_op_iabs:
|
|
|
|
|
case nir_op_fneg:
|
|
|
|
|
case nir_op_fabs:
|
|
|
|
|
case nir_op_fadd:
|
|
|
|
|
case nir_op_iadd:
|
|
|
|
|
case nir_op_iadd_sat:
|
|
|
|
|
case nir_op_uadd_sat:
|
|
|
|
|
case nir_op_isub_sat:
|
|
|
|
|
case nir_op_usub_sat:
|
|
|
|
|
case nir_op_irhadd:
|
|
|
|
|
case nir_op_urhadd:
|
|
|
|
|
case nir_op_fmul:
|
|
|
|
|
case nir_op_inot:
|
|
|
|
|
case nir_op_iand:
|
|
|
|
|
case nir_op_ior:
|
|
|
|
|
case nir_op_ixor:
|
|
|
|
|
case nir_op_ffloor:
|
|
|
|
|
case nir_op_ffract:
|
|
|
|
|
case nir_op_uclz:
|
|
|
|
|
case nir_op_ishl:
|
|
|
|
|
case nir_op_ishr:
|
|
|
|
|
case nir_op_ushr:
|
|
|
|
|
case nir_op_urol:
|
|
|
|
|
case nir_op_uror:
|
|
|
|
|
break; /* ... from switch. */
|
|
|
|
|
default:
|
|
|
|
|
continue; /* ... with loop. */
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* To help prevent increasing live ranges, require that one of the
|
|
|
|
|
* sources be a constant.
|
|
|
|
|
*/
|
|
|
|
|
if (nir_op_infos[alu->op].num_inputs == 2 &&
|
|
|
|
|
!nir_src_is_const(alu->src[0].src) &&
|
|
|
|
|
!nir_src_is_const(alu->src[1].src))
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
if (!all_uses_are_compare_with_zero(alu))
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
/* At this point it is known that the alu is only used by a
|
|
|
|
|
* comparison with zero that is used by nir_op_bcsel and possibly by
|
|
|
|
|
* if-statements (though the latter has not been explicitly checked).
|
|
|
|
|
*
|
|
|
|
|
* Iterate through each use of the ALU. For every use that is in a
|
|
|
|
|
* different block, emit a copy of the ALU. Care must be taken here.
|
|
|
|
|
* The original instruction must be duplicated only once in each
|
|
|
|
|
* block because CSE cannot be run after this pass.
|
|
|
|
|
*/
|
|
|
|
|
nir_foreach_use_safe(use, &alu->def) {
|
2023-08-14 09:58:47 -04:00
|
|
|
nir_instr *const use_instr = nir_src_parent_instr(use);
|
nir/rematerialize: Rematerialize ALUs used only by compares with zero
This was 4th on the list of things to try in 3ee2e84c608 ("nir:
Rematerialize compare instructions"). This is implemented as a separate
subpass that tries to find ALU instructions (with restrictions) that are
only used by comparisons with zero that are in turn only used as
conditions for bcsel or if-statements.
There are two restrictions implemented. One of the sources must be a
constant. This is done in an attempt to prevent increasing register
pressure. Additionally, the opcode of the instruction must be one that
has a high probablility of getting a conditional modifier on Intel
GPUs. Not all instructions can have a conditional modifiers (e.g., min
and max), so I don't think there is any benefit to moving these
instructions.
v2: Rebase on many, many recent NIR infrastructure changes.
v3: Make data in commit message more clear. Suggested by Matt. Rebase on
b5d6b7c402a ("nir: Drop most uses if nir_instr_rewrite_src()").
All of the affected shaders on ILK and G45 are in CS:GO. There is some
brief analysis of the changes in the MR.
Reviewed-by: Matt Tuner <mattst88@gmail.com>
Shader-db results:
DG2
total instructions in shared programs: 22824637 -> 22824258 (<.01%)
instructions in affected programs: 365742 -> 365363 (-0.10%)
helped: 190 / HURT: 97
total cycles in shared programs: 832186193 -> 832157290 (<.01%)
cycles in affected programs: 41245259 -> 41216356 (-0.07%)
helped: 208 / HURT: 117
total spills in shared programs: 4072 -> 4060 (-0.29%)
spills in affected programs: 366 -> 354 (-3.28%)
helped: 4 / HURT: 2
total fills in shared programs: 3601 -> 3607 (0.17%)
fills in affected programs: 708 -> 714 (0.85%)
helped: 4 / HURT: 2
LOST: 0
GAINED: 1
Tiger Lake and Ice Lake had similar results. (Ice Lake shown)
total instructions in shared programs: 20320934 -> 20320689 (<.01%)
instructions in affected programs: 236592 -> 236347 (-0.10%)
helped: 176 / HURT: 29
total cycles in shared programs: 849846341 -> 849843856 (<.01%)
cycles in affected programs: 41277336 -> 41274851 (<.01%)
helped: 195 / HURT: 110
LOST: 0
GAINED: 1
Skylake
total instructions in shared programs: 18550811 -> 18550470 (<.01%)
instructions in affected programs: 233908 -> 233567 (-0.15%)
helped: 182 / HURT: 25
total cycles in shared programs: 835910983 -> 835889167 (<.01%)
cycles in affected programs: 38764359 -> 38742543 (-0.06%)
helped: 207/ HURT: 94
total spills in shared programs: 4522 -> 4506 (-0.35%)
spills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
total fills in shared programs: 5296 -> 5280 (-0.30%)
fills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
LOST: 0
GAINED: 1
Broadwell
total instructions in shared programs: 18199130 -> 18197920 (<.01%)
instructions in affected programs: 214664 -> 213454 (-0.56%)
helped: 191 / HURT: 0
total cycles in shared programs: 935131908 -> 934870248 (-0.03%)
cycles in affected programs: 75770568 -> 75508908 (-0.35%)
helped: 203 / HURT: 84
total spills in shared programs: 13896 -> 13734 (-1.17%)
spills in affected programs: 162 -> 0
helped: 3 / HURT: 0
total fills in shared programs: 16989 -> 16761 (-1.34%)
fills in affected programs: 228 -> 0
helped: 3 / HURT: 0
Haswell
total instructions in shared programs: 16969502 -> 16969085 (<.01%)
instructions in affected programs: 185498 -> 185081 (-0.22%)
helped: 121 / HURT: 1
total cycles in shared programs: 925290863 -> 924806827 (-0.05%)
cycles in affected programs: 30200863 -> 29716827 (-1.60%)
helped: 100 / HURT: 85
total spills in shared programs: 13565 -> 13533 (-0.24%)
spills in affected programs: 736 -> 704 (-4.35%)
helped: 8 / HURT: 0
total fills in shared programs: 15468 -> 15436 (-0.21%)
fills in affected programs: 740 -> 708 (-4.32%)
helped: 8 / HURT: 0
LOST: 0
GAINED: 1
Ivy Bridge
total instructions in shared programs: 15839127 -> 15838947 (<.01%)
instructions in affected programs: 77776 -> 77596 (-0.23%)
helped: 58 / HURT: 0
total cycles in shared programs: 459852774 -> 459739770 (-0.02%)
cycles in affected programs: 11970210 -> 11857206 (-0.94%)
helped: 79 / HURT: 53
Sandy Bridge
total instructions in shared programs: 14106847 -> 14106831 (<.01%)
instructions in affected programs: 1611 -> 1595 (-0.99%)
helped: 10 / HURT: 0
total cycles in shared programs: 775004024 -> 775007516 (<.01%)
cycles in affected programs: 2530686 -> 2534178 (0.14%)
helped: 55 / HURT: 48
Iron Lake
total cycles in shared programs: 257753356 -> 257754900 (<.01%)
cycles in affected programs: 2977374 -> 2978918 (0.05%)
helped: 12 / HURT: 106
GM45
total cycles in shared programs: 169711382 -> 169712816 (<.01%)
cycles in affected programs: 2402070 -> 2403504 (0.06%)
helped: 12 / HURT: 57
Fossil-db results:
All Intel platforms had similar results. (DG2 shown)
Totals:
Instrs: 193884596 -> 193465896 (-0.22%); split: -0.25%, +0.03%
Cycles: 14050193354 -> 14048194826 (-0.01%); split: -0.34%, +0.33%
Spill count: 114944 -> 100449 (-12.61%); split: -13.59%, +0.98%
Fill count: 201525 -> 179534 (-10.91%); split: -11.22%, +0.31%
Scratch Memory Size: 10028032 -> 8468480 (-15.55%)
Totals from 16912 (2.59% of 653124) affected shaders:
Instrs: 34173709 -> 33755009 (-1.23%); split: -1.41%, +0.19%
Cycles: 2945969110 -> 2943970582 (-0.07%); split: -1.62%, +1.55%
Spill count: 97753 -> 83258 (-14.83%); split: -15.98%, +1.15%
Fill count: 176355 -> 154364 (-12.47%); split: -12.82%, +0.35%
Scratch Memory Size: 8619008 -> 7059456 (-18.09%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20176>
2022-12-05 14:18:33 -08:00
|
|
|
|
|
|
|
|
/* If the use is in the same block as the def, don't
|
|
|
|
|
* rematerialize.
|
|
|
|
|
*/
|
|
|
|
|
if (use_instr->block == alu->instr.block)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
nir_alu_instr *clone = nir_alu_instr_clone(shader, alu);
|
|
|
|
|
|
|
|
|
|
nir_instr_insert_before(use_instr, &clone->instr);
|
|
|
|
|
|
|
|
|
|
nir_alu_instr *const use_alu = nir_instr_as_alu(use_instr);
|
|
|
|
|
for (unsigned i = 0; i < nir_op_infos[use_alu->op].num_inputs; i++) {
|
|
|
|
|
if (use_alu->src[i].src.ssa == &alu->def) {
|
|
|
|
|
nir_src_rewrite(&use_alu->src[i].src, &clone->def);
|
|
|
|
|
progress = true;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (progress) {
|
2024-06-16 16:32:01 -04:00
|
|
|
nir_metadata_preserve(impl, nir_metadata_control_flow);
|
nir/rematerialize: Rematerialize ALUs used only by compares with zero
This was 4th on the list of things to try in 3ee2e84c608 ("nir:
Rematerialize compare instructions"). This is implemented as a separate
subpass that tries to find ALU instructions (with restrictions) that are
only used by comparisons with zero that are in turn only used as
conditions for bcsel or if-statements.
There are two restrictions implemented. One of the sources must be a
constant. This is done in an attempt to prevent increasing register
pressure. Additionally, the opcode of the instruction must be one that
has a high probablility of getting a conditional modifier on Intel
GPUs. Not all instructions can have a conditional modifiers (e.g., min
and max), so I don't think there is any benefit to moving these
instructions.
v2: Rebase on many, many recent NIR infrastructure changes.
v3: Make data in commit message more clear. Suggested by Matt. Rebase on
b5d6b7c402a ("nir: Drop most uses if nir_instr_rewrite_src()").
All of the affected shaders on ILK and G45 are in CS:GO. There is some
brief analysis of the changes in the MR.
Reviewed-by: Matt Tuner <mattst88@gmail.com>
Shader-db results:
DG2
total instructions in shared programs: 22824637 -> 22824258 (<.01%)
instructions in affected programs: 365742 -> 365363 (-0.10%)
helped: 190 / HURT: 97
total cycles in shared programs: 832186193 -> 832157290 (<.01%)
cycles in affected programs: 41245259 -> 41216356 (-0.07%)
helped: 208 / HURT: 117
total spills in shared programs: 4072 -> 4060 (-0.29%)
spills in affected programs: 366 -> 354 (-3.28%)
helped: 4 / HURT: 2
total fills in shared programs: 3601 -> 3607 (0.17%)
fills in affected programs: 708 -> 714 (0.85%)
helped: 4 / HURT: 2
LOST: 0
GAINED: 1
Tiger Lake and Ice Lake had similar results. (Ice Lake shown)
total instructions in shared programs: 20320934 -> 20320689 (<.01%)
instructions in affected programs: 236592 -> 236347 (-0.10%)
helped: 176 / HURT: 29
total cycles in shared programs: 849846341 -> 849843856 (<.01%)
cycles in affected programs: 41277336 -> 41274851 (<.01%)
helped: 195 / HURT: 110
LOST: 0
GAINED: 1
Skylake
total instructions in shared programs: 18550811 -> 18550470 (<.01%)
instructions in affected programs: 233908 -> 233567 (-0.15%)
helped: 182 / HURT: 25
total cycles in shared programs: 835910983 -> 835889167 (<.01%)
cycles in affected programs: 38764359 -> 38742543 (-0.06%)
helped: 207/ HURT: 94
total spills in shared programs: 4522 -> 4506 (-0.35%)
spills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
total fills in shared programs: 5296 -> 5280 (-0.30%)
fills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
LOST: 0
GAINED: 1
Broadwell
total instructions in shared programs: 18199130 -> 18197920 (<.01%)
instructions in affected programs: 214664 -> 213454 (-0.56%)
helped: 191 / HURT: 0
total cycles in shared programs: 935131908 -> 934870248 (-0.03%)
cycles in affected programs: 75770568 -> 75508908 (-0.35%)
helped: 203 / HURT: 84
total spills in shared programs: 13896 -> 13734 (-1.17%)
spills in affected programs: 162 -> 0
helped: 3 / HURT: 0
total fills in shared programs: 16989 -> 16761 (-1.34%)
fills in affected programs: 228 -> 0
helped: 3 / HURT: 0
Haswell
total instructions in shared programs: 16969502 -> 16969085 (<.01%)
instructions in affected programs: 185498 -> 185081 (-0.22%)
helped: 121 / HURT: 1
total cycles in shared programs: 925290863 -> 924806827 (-0.05%)
cycles in affected programs: 30200863 -> 29716827 (-1.60%)
helped: 100 / HURT: 85
total spills in shared programs: 13565 -> 13533 (-0.24%)
spills in affected programs: 736 -> 704 (-4.35%)
helped: 8 / HURT: 0
total fills in shared programs: 15468 -> 15436 (-0.21%)
fills in affected programs: 740 -> 708 (-4.32%)
helped: 8 / HURT: 0
LOST: 0
GAINED: 1
Ivy Bridge
total instructions in shared programs: 15839127 -> 15838947 (<.01%)
instructions in affected programs: 77776 -> 77596 (-0.23%)
helped: 58 / HURT: 0
total cycles in shared programs: 459852774 -> 459739770 (-0.02%)
cycles in affected programs: 11970210 -> 11857206 (-0.94%)
helped: 79 / HURT: 53
Sandy Bridge
total instructions in shared programs: 14106847 -> 14106831 (<.01%)
instructions in affected programs: 1611 -> 1595 (-0.99%)
helped: 10 / HURT: 0
total cycles in shared programs: 775004024 -> 775007516 (<.01%)
cycles in affected programs: 2530686 -> 2534178 (0.14%)
helped: 55 / HURT: 48
Iron Lake
total cycles in shared programs: 257753356 -> 257754900 (<.01%)
cycles in affected programs: 2977374 -> 2978918 (0.05%)
helped: 12 / HURT: 106
GM45
total cycles in shared programs: 169711382 -> 169712816 (<.01%)
cycles in affected programs: 2402070 -> 2403504 (0.06%)
helped: 12 / HURT: 57
Fossil-db results:
All Intel platforms had similar results. (DG2 shown)
Totals:
Instrs: 193884596 -> 193465896 (-0.22%); split: -0.25%, +0.03%
Cycles: 14050193354 -> 14048194826 (-0.01%); split: -0.34%, +0.33%
Spill count: 114944 -> 100449 (-12.61%); split: -13.59%, +0.98%
Fill count: 201525 -> 179534 (-10.91%); split: -11.22%, +0.31%
Scratch Memory Size: 10028032 -> 8468480 (-15.55%)
Totals from 16912 (2.59% of 653124) affected shaders:
Instrs: 34173709 -> 33755009 (-1.23%); split: -1.41%, +0.19%
Cycles: 2945969110 -> 2943970582 (-0.07%); split: -1.62%, +1.55%
Spill count: 97753 -> 83258 (-14.83%); split: -15.98%, +1.15%
Fill count: 176355 -> 154364 (-12.47%); split: -12.82%, +0.35%
Scratch Memory Size: 8619008 -> 7059456 (-18.09%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20176>
2022-12-05 14:18:33 -08:00
|
|
|
} else {
|
|
|
|
|
nir_metadata_preserve(impl, nir_metadata_all);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return progress;
|
|
|
|
|
}
|
|
|
|
|
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
bool
|
|
|
|
|
nir_opt_rematerialize_compares(nir_shader *shader)
|
|
|
|
|
{
|
|
|
|
|
bool progress = false;
|
|
|
|
|
|
2023-06-22 13:27:59 -04:00
|
|
|
nir_foreach_function_impl(impl, shader) {
|
|
|
|
|
progress = nir_opt_rematerialize_compares_impl(shader, impl) || progress;
|
nir/rematerialize: Rematerialize ALUs used only by compares with zero
This was 4th on the list of things to try in 3ee2e84c608 ("nir:
Rematerialize compare instructions"). This is implemented as a separate
subpass that tries to find ALU instructions (with restrictions) that are
only used by comparisons with zero that are in turn only used as
conditions for bcsel or if-statements.
There are two restrictions implemented. One of the sources must be a
constant. This is done in an attempt to prevent increasing register
pressure. Additionally, the opcode of the instruction must be one that
has a high probablility of getting a conditional modifier on Intel
GPUs. Not all instructions can have a conditional modifiers (e.g., min
and max), so I don't think there is any benefit to moving these
instructions.
v2: Rebase on many, many recent NIR infrastructure changes.
v3: Make data in commit message more clear. Suggested by Matt. Rebase on
b5d6b7c402a ("nir: Drop most uses if nir_instr_rewrite_src()").
All of the affected shaders on ILK and G45 are in CS:GO. There is some
brief analysis of the changes in the MR.
Reviewed-by: Matt Tuner <mattst88@gmail.com>
Shader-db results:
DG2
total instructions in shared programs: 22824637 -> 22824258 (<.01%)
instructions in affected programs: 365742 -> 365363 (-0.10%)
helped: 190 / HURT: 97
total cycles in shared programs: 832186193 -> 832157290 (<.01%)
cycles in affected programs: 41245259 -> 41216356 (-0.07%)
helped: 208 / HURT: 117
total spills in shared programs: 4072 -> 4060 (-0.29%)
spills in affected programs: 366 -> 354 (-3.28%)
helped: 4 / HURT: 2
total fills in shared programs: 3601 -> 3607 (0.17%)
fills in affected programs: 708 -> 714 (0.85%)
helped: 4 / HURT: 2
LOST: 0
GAINED: 1
Tiger Lake and Ice Lake had similar results. (Ice Lake shown)
total instructions in shared programs: 20320934 -> 20320689 (<.01%)
instructions in affected programs: 236592 -> 236347 (-0.10%)
helped: 176 / HURT: 29
total cycles in shared programs: 849846341 -> 849843856 (<.01%)
cycles in affected programs: 41277336 -> 41274851 (<.01%)
helped: 195 / HURT: 110
LOST: 0
GAINED: 1
Skylake
total instructions in shared programs: 18550811 -> 18550470 (<.01%)
instructions in affected programs: 233908 -> 233567 (-0.15%)
helped: 182 / HURT: 25
total cycles in shared programs: 835910983 -> 835889167 (<.01%)
cycles in affected programs: 38764359 -> 38742543 (-0.06%)
helped: 207/ HURT: 94
total spills in shared programs: 4522 -> 4506 (-0.35%)
spills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
total fills in shared programs: 5296 -> 5280 (-0.30%)
fills in affected programs: 324 -> 308 (-4.94%)
helped: 4 / HURT: 0
LOST: 0
GAINED: 1
Broadwell
total instructions in shared programs: 18199130 -> 18197920 (<.01%)
instructions in affected programs: 214664 -> 213454 (-0.56%)
helped: 191 / HURT: 0
total cycles in shared programs: 935131908 -> 934870248 (-0.03%)
cycles in affected programs: 75770568 -> 75508908 (-0.35%)
helped: 203 / HURT: 84
total spills in shared programs: 13896 -> 13734 (-1.17%)
spills in affected programs: 162 -> 0
helped: 3 / HURT: 0
total fills in shared programs: 16989 -> 16761 (-1.34%)
fills in affected programs: 228 -> 0
helped: 3 / HURT: 0
Haswell
total instructions in shared programs: 16969502 -> 16969085 (<.01%)
instructions in affected programs: 185498 -> 185081 (-0.22%)
helped: 121 / HURT: 1
total cycles in shared programs: 925290863 -> 924806827 (-0.05%)
cycles in affected programs: 30200863 -> 29716827 (-1.60%)
helped: 100 / HURT: 85
total spills in shared programs: 13565 -> 13533 (-0.24%)
spills in affected programs: 736 -> 704 (-4.35%)
helped: 8 / HURT: 0
total fills in shared programs: 15468 -> 15436 (-0.21%)
fills in affected programs: 740 -> 708 (-4.32%)
helped: 8 / HURT: 0
LOST: 0
GAINED: 1
Ivy Bridge
total instructions in shared programs: 15839127 -> 15838947 (<.01%)
instructions in affected programs: 77776 -> 77596 (-0.23%)
helped: 58 / HURT: 0
total cycles in shared programs: 459852774 -> 459739770 (-0.02%)
cycles in affected programs: 11970210 -> 11857206 (-0.94%)
helped: 79 / HURT: 53
Sandy Bridge
total instructions in shared programs: 14106847 -> 14106831 (<.01%)
instructions in affected programs: 1611 -> 1595 (-0.99%)
helped: 10 / HURT: 0
total cycles in shared programs: 775004024 -> 775007516 (<.01%)
cycles in affected programs: 2530686 -> 2534178 (0.14%)
helped: 55 / HURT: 48
Iron Lake
total cycles in shared programs: 257753356 -> 257754900 (<.01%)
cycles in affected programs: 2977374 -> 2978918 (0.05%)
helped: 12 / HURT: 106
GM45
total cycles in shared programs: 169711382 -> 169712816 (<.01%)
cycles in affected programs: 2402070 -> 2403504 (0.06%)
helped: 12 / HURT: 57
Fossil-db results:
All Intel platforms had similar results. (DG2 shown)
Totals:
Instrs: 193884596 -> 193465896 (-0.22%); split: -0.25%, +0.03%
Cycles: 14050193354 -> 14048194826 (-0.01%); split: -0.34%, +0.33%
Spill count: 114944 -> 100449 (-12.61%); split: -13.59%, +0.98%
Fill count: 201525 -> 179534 (-10.91%); split: -11.22%, +0.31%
Scratch Memory Size: 10028032 -> 8468480 (-15.55%)
Totals from 16912 (2.59% of 653124) affected shaders:
Instrs: 34173709 -> 33755009 (-1.23%); split: -1.41%, +0.19%
Cycles: 2945969110 -> 2943970582 (-0.07%); split: -1.62%, +1.55%
Spill count: 97753 -> 83258 (-14.83%); split: -15.98%, +1.15%
Fill count: 176355 -> 154364 (-12.47%); split: -12.82%, +0.35%
Scratch Memory Size: 8619008 -> 7059456 (-18.09%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20176>
2022-12-05 14:18:33 -08:00
|
|
|
|
|
|
|
|
progress = nir_opt_rematerialize_alu_impl(shader, impl) || progress;
|
nir: Rematerialize compare instructions
On some architectures, Boolean values used to control conditional
branches or condtional selection must be propagated into a flag. This
generally means that a stored Boolean value must be compared with zero.
Rather than force the generation of extra compares with zero, re-emit
the original comparison instruction. This can save register pressure by
not needing to store the Boolean value.
There are several possible ares for future improvement to this pass:
1. Be more conservative. If both sources to the comparison instruction
are non-constants, it may be better for register pressure to emit the
extra compare. The current shader-db results on Intel GPUs (next
commit) lead me to believe that this is not currently a problem.
2. Be less conservative. Currently the pass requires that all users of
the comparison match the pattern. The idea is that after the pass is
complete, no instruction will use the resulting Boolean value. The only
uses will be of the flag value. It may be beneficial to relax this
requirement in some cases.
3. Be less conservative. Also try to rematerialize comparisons used for
discard_if intrinsics. After changing the way the Intel compiler
generates cod e for discard_if (see MR!935), I tried implementing this
already. The changes were pretty small. Instructions were helped in 19
shaders, but, overall, cycles were hurt. A commit "nir: Rematerialize
comparisons for nir_intrinsic_discard_if too" is on my fd.o cgit.
4. Copy the preceeding ALU instruction. If the comparison is a
comparison with zero, and it is the only user of a particular ALU
instruction (e.g., (a+b) != 0.0), it may be a further improvment to also
copy the preceeding ALU instruction. On Intel GPUs, this may enable
cmod propagation to make additional progress.
v2: Use much simpler method to get the prev_block for an if-statement.
Suggested by Tim.
Reviewed-by: Matt Turner <mattst88@gmail.com>
2019-05-20 11:22:12 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return progress;
|
|
|
|
|
}
|