mesa/src/gallium/drivers/r600/sfn/sfn_ra.cpp

295 lines
8.9 KiB
C++

/* -*- mesa-c++ -*-
*
* Copyright (c) 2022 Collabora LTD
*
* Author: Gert Wollny <gert.wollny@collabora.com>
*
* 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
* on the rights to use, copy, modify, merge, publish, distribute, sub
* license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
* THE AUTHOR(S) AND/OR THEIR SUPPLIERS 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 "sfn_debug.h"
#include "sfn_ra.h"
#include <cassert>
#include <queue>
namespace r600 {
void ComponentInterference::prepare_row(int row)
{
m_rows.resize(row + 1);
}
void ComponentInterference::add(size_t idx1, size_t idx2)
{
assert(idx1 > idx2);
assert(m_rows.size() > idx1);
m_rows[idx1].push_back(idx2);
m_rows[idx2].push_back(idx1);
}
Interference::Interference(LiveRangeMap& map):
m_map(map)
{
initialize();
}
void Interference::initialize()
{
for(int i = 0; i < 4; ++i) {
initialize(m_components_maps[i], m_map.component(i));
}
}
void Interference::initialize(ComponentInterference& comp_interference,
LiveRangeMap::ChannelLiveRange& clr)
{
for (size_t row = 0; row < clr.size(); ++row) {
auto& row_entry = clr[row];
comp_interference.prepare_row(row);
for (size_t col = 0; col < row; ++col) {
auto& col_entry = clr[col];
if (row_entry.m_end >= col_entry.m_start &&
row_entry.m_start <= col_entry.m_end)
comp_interference.add(row, col);
}
}
}
struct Group {
int priority;
std::array<PRegister, 4> channels;
};
static inline bool operator < (const Group& lhs, const Group& rhs)
{
return lhs.priority < rhs.priority;
}
using GroupRegisters = std::priority_queue<Group>;
static bool
group_allocation (LiveRangeMap& lrm, const Interference& interference, GroupRegisters& groups)
{
int color = 0;
// allocate grouped registers
while (!groups.empty()) {
auto group = groups.top();
groups.pop();
int start_comp = 0;
while (!group.channels[start_comp])
++start_comp;
sfn_log << SfnLog::merge << "Color group with " << *group.channels[start_comp] << "\n";
// don't restart registers for exports, we may be able tp merge the
// export calls, is fthe registers are consecutive
if (group.priority > 0)
color = 0;
while (color < 124) {
/* Find the coloring for the first channel */
bool color_in_use = false;
int comp = start_comp;
auto& adjecency = interference.row(start_comp, group.channels[comp]->index());
auto& regs = lrm.component(comp);
sfn_log << SfnLog::merge << "Try color "<< color;
for (auto adj : adjecency) {
if (regs[adj].m_color == color) {
color_in_use = true;
sfn_log << SfnLog::merge << " in use\n";
break;
}
}
if (color_in_use) {
++color;
continue;
}
/* First channel color found, check whether it can be used for all channels */
while (comp < 4) {
sfn_log << SfnLog::merge << " interference: ";
if (group.channels[comp]) {
auto& component_life_ranges = lrm.component(comp);
auto& adjecencies = interference.row(comp, group.channels[comp]->index());
for (auto adj_index : adjecencies) {
sfn_log << SfnLog::merge << *component_life_ranges[adj_index].m_register << " ";
if (component_life_ranges[adj_index].m_color == color) {
color_in_use = true;
sfn_log << SfnLog::merge << "used";
break;
}
}
if (color_in_use)
break;
}
++comp;
}
/* We couldn't allocate all channels with this color, so try next */
if (color_in_use) {
++color;
sfn_log << SfnLog::merge << "\n";
continue;
}
sfn_log << SfnLog::merge << " success\n";
/* Coloring successful */
for (auto reg : group.channels) {
if (reg) {
auto& vregs = lrm.component(reg->chan());
auto& vreg_cmp = vregs[reg->index()];
assert(vreg_cmp.m_start != -1 || vreg_cmp.m_end != -1);
vreg_cmp.m_color = color;
}
}
break;
}
if (color == 124)
return false;
}
return true;
}
static bool
scalar_allocation (LiveRangeMap& lrm, const Interference& interference)
{
for (int comp = 0; comp < 4; ++comp) {
auto& live_ranges = lrm.component(comp);
for (auto& r : live_ranges) {
if (r.m_color != -1)
continue;
if (r.m_start == -1 &&
r.m_end == -1)
continue;
sfn_log << SfnLog::merge << "Color " << *r.m_register << "\n";
auto& adjecency = interference.row(comp, r.m_register->index());
int color = 0;
while (color < 124) {
bool color_in_use = false;
for (auto adj : adjecency) {
if (live_ranges[adj].m_color == color) {
color_in_use = true;
break;
}
}
if (color_in_use) {
++color;
continue;
}
r.m_color = color;
break;
}
if (color == 124)
return false;
}
}
return true;
}
bool register_allocation(LiveRangeMap& lrm)
{
Interference interference(lrm);
std::map<int, Group> groups;
// setup fixed colors and group relationships
for (int i = 0; i < 4; ++i) {
auto& comp = lrm.component(i);
for (auto& entry : comp) {
sfn_log << SfnLog::merge << "Prepare RA for "
<< *entry.m_register
<< " [" << entry.m_start << ", " << entry.m_end << "]\n";
auto pin = entry.m_register->pin();
if (entry.m_start == -1 && entry.m_end == -1) {
if (pin == pin_group || pin == pin_chgr)
entry.m_register->set_chan(7);
continue;
}
auto sel = entry.m_register->sel();
/* fully pinned registers contain system values with the
* definite register index, and array values are allocated
* right after the system registers, so just reuse the IDs (for now) */
if (pin == pin_fully || pin == pin_array) {
/* Must set all array element entries */
sfn_log << SfnLog::merge << "Pin color " << sel << " to " << *entry.m_register << "\n";
entry.m_color = sel;
} else if (pin == pin_group || pin == pin_chgr) {
/* Groups must all have the same sel() value, because they are used
* as vec4 registers */
auto igroup = groups.find(sel);
if (igroup != groups.end()) {
igroup->second.channels[i] = entry.m_register;
assert(comp[entry.m_register->index()].m_register->index() == entry.m_register->index());
} else {
int priority = entry.m_use.test(LiveRangeEntry::use_export) ? - entry.m_end : entry.m_start;
Group group{priority, {nullptr, nullptr, nullptr, nullptr}};
group.channels[i] = entry.m_register;
assert(comp[group.channels[i]->index()].m_register->index() == entry.m_register->index());
groups[sel] = group;
}
}
}
}
GroupRegisters groups_sorted;
for (auto& [sel, group] : groups)
groups_sorted.push(group);
if (!group_allocation (lrm, interference, groups_sorted))
return false;
if (!scalar_allocation(lrm, interference))
return false;
for (int i = 0; i < 4; ++i) {
auto& comp = lrm.component(i);
for (auto& entry : comp) {
sfn_log << SfnLog::merge << "Set " << *entry.m_register << " to ";
entry.m_register->set_sel(entry.m_color);
entry.m_register->set_pin(pin_none);
sfn_log << SfnLog::merge << *entry.m_register << "\n";
}
}
return true;
}
}