mesa/src/intel/compiler/brw_fs_live_variables.cpp
Ian Romanick 38807ceeae intel/fs: sel.cond writes the flags on Gfx4 and Gfx5
On Gfx4 and Gfx5, sel.l (for min) and sel.ge (for max) are implemented
using a separte cmpn and sel instruction.  This lowering occurs in
fs_vistor::lower_minmax which is called very, very late... a long, long
time after the first calls to opt_cmod_propagation.  As a result,
conditional modifiers can be incorrectly propagated across sel.cond on
those platforms.

No tests were affected by this change, and I find that quite shocking.
After just changing flags_written(), all of the atan tests started
failing on ILK.  That required the change in cmod_propagatin (and the
addition of the prop_across_into_sel_gfx5 unit test).

Shader-db results for ILK and GM45 are below.  I looked at a couple
before and after shaders... and every case that I looked at had
experienced incorrect cmod propagation.  This affected a LOT of apps!
Euro Truck Simulator 2, The Talos Principle, Serious Sam 3, Sanctum 2,
Gang Beasts, and on and on... :(

I discovered this bug while working on a couple new optimization
passes.  One of the passes attempts to remove condition modifiers that
are never used.  The pass made no progress except on ILK and GM45.
After investigating a couple of the affected shaders, I noticed that
the code in those shaders looked wrong... investigation led to this
cause.

v2: Trivial changes in the unit tests.

v3: Fix type in comment in unit tests.  Noticed by Jason and Priit.

v4: Tweak handling of BRW_OPCODE_SEL special case.  Suggested by Jason.

Fixes: df1aec763e ("i965/fs: Define methods to calculate the flag subset read or written by an fs_inst.")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Dave Airlie <airlied@redhat.com>

Iron Lake
total instructions in shared programs: 8180493 -> 8181781 (0.02%)
instructions in affected programs: 541796 -> 543084 (0.24%)
helped: 28
HURT: 1158
helped stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1
helped stats (rel) min: 0.35% max: 0.86% x̄: 0.53% x̃: 0.50%
HURT stats (abs)   min: 1 max: 3 x̄: 1.14 x̃: 1
HURT stats (rel)   min: 0.12% max: 4.00% x̄: 0.37% x̃: 0.23%
95% mean confidence interval for instructions value: 1.06 1.11
95% mean confidence interval for instructions %-change: 0.31% 0.38%
Instructions are HURT.

total cycles in shared programs: 239420470 -> 239421690 (<.01%)
cycles in affected programs: 2925992 -> 2927212 (0.04%)
helped: 49
HURT: 157
helped stats (abs) min: 2 max: 284 x̄: 62.69 x̃: 70
helped stats (rel) min: 0.04% max: 6.20% x̄: 1.68% x̃: 1.96%
HURT stats (abs)   min: 2 max: 48 x̄: 27.34 x̃: 24
HURT stats (rel)   min: 0.02% max: 2.91% x̄: 0.31% x̃: 0.20%
95% mean confidence interval for cycles value: -0.80 12.64
95% mean confidence interval for cycles %-change: -0.31% <.01%
Inconclusive result (value mean confidence interval includes 0).

GM45
total instructions in shared programs: 4985517 -> 4986207 (0.01%)
instructions in affected programs: 306935 -> 307625 (0.22%)
helped: 14
HURT: 625
helped stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1
helped stats (rel) min: 0.35% max: 0.82% x̄: 0.52% x̃: 0.49%
HURT stats (abs)   min: 1 max: 3 x̄: 1.13 x̃: 1
HURT stats (rel)   min: 0.12% max: 3.90% x̄: 0.34% x̃: 0.22%
95% mean confidence interval for instructions value: 1.04 1.12
95% mean confidence interval for instructions %-change: 0.29% 0.36%
Instructions are HURT.

total cycles in shared programs: 153827268 -> 153828052 (<.01%)
cycles in affected programs: 1669290 -> 1670074 (0.05%)
helped: 24
HURT: 84
helped stats (abs) min: 2 max: 232 x̄: 64.33 x̃: 67
helped stats (rel) min: 0.04% max: 4.62% x̄: 1.60% x̃: 1.94%
HURT stats (abs)   min: 2 max: 48 x̄: 27.71 x̃: 24
HURT stats (rel)   min: 0.02% max: 2.66% x̄: 0.34% x̃: 0.14%
95% mean confidence interval for cycles value: -1.94 16.46
95% mean confidence interval for cycles %-change: -0.29% 0.11%
Inconclusive result (value mean confidence interval includes 0).

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12191>
2021-08-11 13:09:20 -07:00

379 lines
12 KiB
C++

/*
* Copyright © 2012 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:
* Eric Anholt <eric@anholt.net>
*
*/
#include "brw_fs.h"
#include "brw_fs_live_variables.h"
using namespace brw;
#define MAX_INSTRUCTION (1 << 30)
/** @file brw_fs_live_variables.cpp
*
* Support for calculating liveness information about virtual GRFs.
*
* This produces a live interval for each whole virtual GRF. We could
* choose to expose per-component live intervals for VGRFs of size > 1,
* but we currently do not. It is easier for the consumers of this
* information to work with whole VGRFs.
*
* However, we internally track use/def information at the per-GRF level for
* greater accuracy. Large VGRFs may be accessed piecemeal over many
* (possibly non-adjacent) instructions. In this case, examining a single
* instruction is insufficient to decide whether a whole VGRF is ultimately
* used or defined. Tracking individual components allows us to easily
* assemble this information.
*
* See Muchnick's Advanced Compiler Design and Implementation, section
* 14.1 (p444).
*/
void
fs_live_variables::setup_one_read(struct block_data *bd,
int ip, const fs_reg &reg)
{
int var = var_from_reg(reg);
assert(var < num_vars);
start[var] = MIN2(start[var], ip);
end[var] = MAX2(end[var], ip);
/* The use[] bitset marks when the block makes use of a variable (VGRF
* channel) without having completely defined that variable within the
* block.
*/
if (!BITSET_TEST(bd->def, var))
BITSET_SET(bd->use, var);
}
void
fs_live_variables::setup_one_write(struct block_data *bd, fs_inst *inst,
int ip, const fs_reg &reg)
{
int var = var_from_reg(reg);
assert(var < num_vars);
start[var] = MIN2(start[var], ip);
end[var] = MAX2(end[var], ip);
/* The def[] bitset marks when an initialization in a block completely
* screens off previous updates of that variable (VGRF channel).
*/
if (inst->dst.file == VGRF) {
if (!inst->is_partial_write() && !BITSET_TEST(bd->use, var))
BITSET_SET(bd->def, var);
BITSET_SET(bd->defout, var);
}
}
/**
* Sets up the use[] and def[] bitsets.
*
* The basic-block-level live variable analysis needs to know which
* variables get used before they're completely defined, and which
* variables are completely defined before they're used.
*
* These are tracked at the per-component level, rather than whole VGRFs.
*/
void
fs_live_variables::setup_def_use()
{
int ip = 0;
foreach_block (block, cfg) {
assert(ip == block->start_ip);
if (block->num > 0)
assert(cfg->blocks[block->num - 1]->end_ip == ip - 1);
struct block_data *bd = &block_data[block->num];
foreach_inst_in_block(fs_inst, inst, block) {
/* Set use[] for this instruction */
for (unsigned int i = 0; i < inst->sources; i++) {
fs_reg reg = inst->src[i];
if (reg.file != VGRF)
continue;
for (unsigned j = 0; j < regs_read(inst, i); j++) {
setup_one_read(bd, ip, reg);
reg.offset += REG_SIZE;
}
}
bd->flag_use[0] |= inst->flags_read(devinfo) & ~bd->flag_def[0];
/* Set def[] for this instruction */
if (inst->dst.file == VGRF) {
fs_reg reg = inst->dst;
for (unsigned j = 0; j < regs_written(inst); j++) {
setup_one_write(bd, inst, ip, reg);
reg.offset += REG_SIZE;
}
}
if (!inst->predicate && inst->exec_size >= 8)
bd->flag_def[0] |= inst->flags_written(devinfo) & ~bd->flag_use[0];
ip++;
}
}
}
/**
* The algorithm incrementally sets bits in liveout and livein,
* propagating it through control flow. It will eventually terminate
* because it only ever adds bits, and stops when no bits are added in
* a pass.
*/
void
fs_live_variables::compute_live_variables()
{
bool cont = true;
while (cont) {
cont = false;
foreach_block_reverse (block, cfg) {
struct block_data *bd = &block_data[block->num];
/* Update liveout */
foreach_list_typed(bblock_link, child_link, link, &block->children) {
struct block_data *child_bd = &block_data[child_link->block->num];
for (int i = 0; i < bitset_words; i++) {
BITSET_WORD new_liveout = (child_bd->livein[i] &
~bd->liveout[i]);
if (new_liveout) {
bd->liveout[i] |= new_liveout;
cont = true;
}
}
BITSET_WORD new_liveout = (child_bd->flag_livein[0] &
~bd->flag_liveout[0]);
if (new_liveout) {
bd->flag_liveout[0] |= new_liveout;
cont = true;
}
}
/* Update livein */
for (int i = 0; i < bitset_words; i++) {
BITSET_WORD new_livein = (bd->use[i] |
(bd->liveout[i] &
~bd->def[i]));
if (new_livein & ~bd->livein[i]) {
bd->livein[i] |= new_livein;
cont = true;
}
}
BITSET_WORD new_livein = (bd->flag_use[0] |
(bd->flag_liveout[0] &
~bd->flag_def[0]));
if (new_livein & ~bd->flag_livein[0]) {
bd->flag_livein[0] |= new_livein;
cont = true;
}
}
}
/* Propagate defin and defout down the CFG to calculate the union of live
* variables potentially defined along any possible control flow path.
*/
do {
cont = false;
foreach_block (block, cfg) {
const struct block_data *bd = &block_data[block->num];
foreach_list_typed(bblock_link, child_link, link, &block->children) {
struct block_data *child_bd = &block_data[child_link->block->num];
for (int i = 0; i < bitset_words; i++) {
const BITSET_WORD new_def = bd->defout[i] & ~child_bd->defin[i];
child_bd->defin[i] |= new_def;
child_bd->defout[i] |= new_def;
cont |= new_def;
}
}
}
} while (cont);
}
/**
* Extend the start/end ranges for each variable to account for the
* new information calculated from control flow.
*/
void
fs_live_variables::compute_start_end()
{
foreach_block (block, cfg) {
struct block_data *bd = &block_data[block->num];
for (int w = 0; w < bitset_words; w++) {
BITSET_WORD livedefin = bd->livein[w] & bd->defin[w];
BITSET_WORD livedefout = bd->liveout[w] & bd->defout[w];
BITSET_WORD livedefinout = livedefin | livedefout;
while (livedefinout) {
unsigned b = u_bit_scan(&livedefinout);
unsigned i = w * BITSET_WORDBITS + b;
if (livedefin & (1u << b)) {
start[i] = MIN2(start[i], block->start_ip);
end[i] = MAX2(end[i], block->start_ip);
}
if (livedefout & (1u << b)) {
start[i] = MIN2(start[i], block->end_ip);
end[i] = MAX2(end[i], block->end_ip);
}
}
}
}
}
fs_live_variables::fs_live_variables(const backend_shader *s)
: devinfo(s->devinfo), cfg(s->cfg)
{
mem_ctx = ralloc_context(NULL);
num_vgrfs = s->alloc.count;
num_vars = 0;
var_from_vgrf = rzalloc_array(mem_ctx, int, num_vgrfs);
for (int i = 0; i < num_vgrfs; i++) {
var_from_vgrf[i] = num_vars;
num_vars += s->alloc.sizes[i];
}
vgrf_from_var = rzalloc_array(mem_ctx, int, num_vars);
for (int i = 0; i < num_vgrfs; i++) {
for (unsigned j = 0; j < s->alloc.sizes[i]; j++) {
vgrf_from_var[var_from_vgrf[i] + j] = i;
}
}
start = ralloc_array(mem_ctx, int, num_vars);
end = rzalloc_array(mem_ctx, int, num_vars);
for (int i = 0; i < num_vars; i++) {
start[i] = MAX_INSTRUCTION;
end[i] = -1;
}
vgrf_start = ralloc_array(mem_ctx, int, num_vgrfs);
vgrf_end = ralloc_array(mem_ctx, int, num_vgrfs);
for (int i = 0; i < num_vgrfs; i++) {
vgrf_start[i] = MAX_INSTRUCTION;
vgrf_end[i] = -1;
}
block_data = rzalloc_array(mem_ctx, struct block_data, cfg->num_blocks);
bitset_words = BITSET_WORDS(num_vars);
for (int i = 0; i < cfg->num_blocks; i++) {
block_data[i].def = rzalloc_array(mem_ctx, BITSET_WORD, bitset_words);
block_data[i].use = rzalloc_array(mem_ctx, BITSET_WORD, bitset_words);
block_data[i].livein = rzalloc_array(mem_ctx, BITSET_WORD, bitset_words);
block_data[i].liveout = rzalloc_array(mem_ctx, BITSET_WORD, bitset_words);
block_data[i].defin = rzalloc_array(mem_ctx, BITSET_WORD, bitset_words);
block_data[i].defout = rzalloc_array(mem_ctx, BITSET_WORD, bitset_words);
block_data[i].flag_def[0] = 0;
block_data[i].flag_use[0] = 0;
block_data[i].flag_livein[0] = 0;
block_data[i].flag_liveout[0] = 0;
}
setup_def_use();
compute_live_variables();
compute_start_end();
/* Merge the per-component live ranges to whole VGRF live ranges. */
for (int i = 0; i < num_vars; i++) {
const unsigned vgrf = vgrf_from_var[i];
vgrf_start[vgrf] = MIN2(vgrf_start[vgrf], start[i]);
vgrf_end[vgrf] = MAX2(vgrf_end[vgrf], end[i]);
}
}
fs_live_variables::~fs_live_variables()
{
ralloc_free(mem_ctx);
}
static bool
check_register_live_range(const fs_live_variables *live, int ip,
const fs_reg &reg, unsigned n)
{
const unsigned var = live->var_from_reg(reg);
if (var + n > unsigned(live->num_vars) ||
live->vgrf_start[reg.nr] > ip || live->vgrf_end[reg.nr] < ip)
return false;
for (unsigned j = 0; j < n; j++) {
if (live->start[var + j] > ip || live->end[var + j] < ip)
return false;
}
return true;
}
bool
fs_live_variables::validate(const backend_shader *s) const
{
int ip = 0;
foreach_block_and_inst(block, fs_inst, inst, s->cfg) {
for (unsigned i = 0; i < inst->sources; i++) {
if (inst->src[i].file == VGRF &&
!check_register_live_range(this, ip,
inst->src[i], regs_read(inst, i)))
return false;
}
if (inst->dst.file == VGRF &&
!check_register_live_range(this, ip, inst->dst, regs_written(inst)))
return false;
ip++;
}
return true;
}
bool
fs_live_variables::vars_interfere(int a, int b) const
{
return !(end[b] <= start[a] ||
end[a] <= start[b]);
}
bool
fs_live_variables::vgrfs_interfere(int a, int b) const
{
return !(vgrf_end[a] <= vgrf_start[b] ||
vgrf_end[b] <= vgrf_start[a]);
}