panfrost/util/lcra.c

/*
 * Copyright (C) 2019 Collabora, Ltd.
 *
 * 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 (Collabora):
 *      Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
 */

#include <stdio.h>
#include <assert.h>
#include <stdlib.h>
#include <string.h>
#include <limits.h>
#include "util/macros.h"
#include "util/u_math.h"
#include "lcra.h"

/* This module is the reference implementation of "Linearly Constrained
 * Register Allocation". The paper is available in PDF form
 * (https://people.collabora.com/~alyssa/LCRA.pdf) as well as Markdown+LaTeX
 * (https://gitlab.freedesktop.org/alyssa/lcra/blob/master/LCRA.md)
 */

struct lcra_state *
lcra_alloc_equations(
                unsigned node_count, unsigned class_count)
{
        struct lcra_state *l = calloc(1, sizeof(*l));

        l->node_count = node_count;
        l->class_count = class_count;

        l->alignment = calloc(sizeof(l->alignment[0]), node_count);
        l->linear = calloc(sizeof(l->linear[0]), node_count * node_count);
        l->modulus = calloc(sizeof(l->modulus[0]), node_count);
        l->class = calloc(sizeof(l->class[0]), node_count);
        l->class_start = calloc(sizeof(l->class_start[0]), class_count);
        l->class_disjoint = calloc(sizeof(l->class_disjoint[0]), class_count * class_count);
        l->class_size = calloc(sizeof(l->class_size[0]), class_count);
        l->spill_cost = calloc(sizeof(l->spill_cost[0]), node_count);
        l->solutions = calloc(sizeof(l->solutions[0]), node_count);

        memset(l->solutions, ~0, sizeof(l->solutions[0]) * node_count);

        return l;
}

void
lcra_free(struct lcra_state *l)
{
        if (!l)
                return;

        free(l->alignment);
        free(l->linear);
        free(l->modulus);
        free(l->class);
        free(l->class_start);
        free(l->class_disjoint);
        free(l->class_size);
        free(l->spill_cost);
        free(l->solutions);

        free(l);
}

void
lcra_set_alignment(struct lcra_state *l, unsigned node, unsigned align_log2, unsigned bound)
{
        l->alignment[node] = (align_log2 + 1) | (bound << 16);
}

void
lcra_set_disjoint_class(struct lcra_state *l, unsigned c1, unsigned c2)
{
        l->class_disjoint[(c1 * l->class_count) + c2] = true;
        l->class_disjoint[(c2 * l->class_count) + c1] = true;
}

void
lcra_restrict_range(struct lcra_state *l, unsigned node, unsigned len)
{
        if (node < l->node_count && l->alignment[node]) {
                unsigned BA = l->alignment[node];
                unsigned alignment = (BA & 0xffff) - 1;
                unsigned bound = BA >> 16;
                l->modulus[node] = DIV_ROUND_UP(bound - len + 1, 1 << alignment);
        }
}

void
lcra_add_node_interference(struct lcra_state *l, unsigned i, unsigned cmask_i, unsigned j, unsigned cmask_j)
{
        if (i == j)
                return;

        if (l->class_disjoint[(l->class[i] * l->class_count) + l->class[j]])
                return;

        uint32_t constraint_fw = 0;
        uint32_t constraint_bw = 0;

        for (unsigned D = 0; D < 16; ++D) {
                if (cmask_i & (cmask_j << D)) {
                        constraint_bw |= (1 << (15 + D));
                        constraint_fw |= (1 << (15 - D));
                }

                if (cmask_i & (cmask_j >> D)) {
                        constraint_fw |= (1 << (15 + D));
                        constraint_bw |= (1 << (15 - D));
                }
        }

        l->linear[j * l->node_count + i] |= constraint_fw;
        l->linear[i * l->node_count + j] |= constraint_bw;
}

static bool
lcra_test_linear(struct lcra_state *l, unsigned *solutions, unsigned i)
{
        unsigned *row = &l->linear[i * l->node_count];
        signed constant = solutions[i];

        for (unsigned j = 0; j < l->node_count; ++j) {
                if (solutions[j] == ~0) continue;

                signed lhs = solutions[j] - constant;

                if (lhs < -15 || lhs > 15)
                        continue;

                if (row[j] & (1 << (lhs + 15)))
                        return false;
        }

        return true;
}

bool
lcra_solve(struct lcra_state *l)
{
        for (unsigned step = 0; step < l->node_count; ++step) {
                if (l->solutions[step] != ~0) continue;
                if (l->alignment[step] == 0) continue;

                unsigned _class = l->class[step];
                unsigned class_start = l->class_start[_class];

                unsigned BA = l->alignment[step];
                unsigned shift = (BA & 0xffff) - 1;
                unsigned bound = BA >> 16;

                unsigned P = bound >> shift;
                unsigned Q = l->modulus[step];
                unsigned r_max = l->class_size[_class];
                unsigned k_max = r_max >> shift;
                unsigned m_max = k_max / P;
                bool succ = false;

                for (unsigned m = 0; m < m_max; ++m) {
                        for (unsigned n = 0; n < Q; ++n) {
                                l->solutions[step] = ((m * P + n) << shift) + class_start;
                                succ = lcra_test_linear(l, l->solutions, step);

                                if (succ) break;
                        }

                        if (succ) break;
                }

                /* Out of registers - prepare to spill */
                if (!succ) {
                        l->spill_class = l->class[step];
                        return false;
                }
        }

        return true;
}

/* Register spilling is implemented with a cost-benefit system. Costs are set
 * by the user. Benefits are calculated from the constraints. */

void
lcra_set_node_spill_cost(struct lcra_state *l, unsigned node, signed cost)
{
        if (node < l->node_count)
                l->spill_cost[node] = cost;
}

/* Count along the lower triangle */

static unsigned
lcra_count_constraints(struct lcra_state *l, unsigned i)
{
        unsigned count = 0;
        unsigned *constraints = &l->linear[i * l->node_count];

        for (unsigned j = 0; j < i; ++j)
                count += util_bitcount(constraints[j]);

        return count;
}

signed
lcra_get_best_spill_node(struct lcra_state *l)
{
        float best_benefit = -1.0;
        signed best_node = -1;

        for (unsigned i = 0; i < l->node_count; ++i) {
                /* Find spillable nodes */
                if (l->class[i] != l->spill_class) continue;
                if (l->spill_cost[i] < 0) continue;

                /* Adapted from Chaitin's heuristic */
                float constraints = lcra_count_constraints(l, i);
                float cost = (l->spill_cost[i] + 1);
                float benefit = constraints / cost;

                if (benefit > best_benefit) {
                        best_benefit = benefit;
                        best_node = i;
                }
        }

        return best_node;
}