mesa/src/panfrost/bifrost/bifrost_compile.c

/*
 * Copyright (C) 2020 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 "main/mtypes.h"
#include "compiler/glsl/glsl_to_nir.h"
#include "compiler/nir_types.h"
#include "main/imports.h"
#include "compiler/nir/nir_builder.h"

#include "disassemble.h"
#include "bifrost_compile.h"
#include "compiler.h"
#include "bi_quirks.h"
#include "bi_print.h"

static bi_block *emit_cf_list(bi_context *ctx, struct exec_list *list);

static bi_block *
create_empty_block(bi_context *ctx)
{
        bi_block *blk = rzalloc(ctx, bi_block);

        blk->predecessors = _mesa_set_create(blk,
                        _mesa_hash_pointer,
                        _mesa_key_pointer_equal);

        blk->name = ctx->block_name_count++;

        return blk;
}

static bi_block *
emit_block(bi_context *ctx, nir_block *block)
{
        ctx->current_block = create_empty_block(ctx);
        list_addtail(&ctx->current_block->link, &ctx->blocks);
        list_inithead(&ctx->current_block->instructions);

        nir_foreach_instr(instr, block) {
                //emit_instr(ctx, instr);
                ++ctx->instruction_count;
        }

        return ctx->current_block;
}

static bi_block *
emit_cf_list(bi_context *ctx, struct exec_list *list)
{
        bi_block *start_block = NULL;

        foreach_list_typed(nir_cf_node, node, node, list) {
                switch (node->type) {
                case nir_cf_node_block: {
                        bi_block *block = emit_block(ctx, nir_cf_node_as_block(node));

                        if (!start_block)
                                start_block = block;

                        break;
                }

#if 0
                case nir_cf_node_if:
                        emit_if(ctx, nir_cf_node_as_if(node));
                        break;

                case nir_cf_node_loop:
                        emit_loop(ctx, nir_cf_node_as_loop(node));
                        break;
#endif

                default:
                        unreachable("Unknown control flow");
                }
        }

        return start_block;
}

static int
glsl_type_size(const struct glsl_type *type, bool bindless)
{
        return glsl_count_attribute_slots(type, false);
}

static void
bi_optimize_nir(nir_shader *nir)
{
        bool progress;
        unsigned lower_flrp = 16 | 32 | 64;

        NIR_PASS(progress, nir, nir_lower_regs_to_ssa);
        NIR_PASS(progress, nir, nir_lower_idiv, nir_lower_idiv_fast);

        nir_lower_tex_options lower_tex_options = {
                .lower_txs_lod = true,
                .lower_txp = ~0,
                .lower_tex_without_implicit_lod = true,
                .lower_txd = true,
        };

        NIR_PASS(progress, nir, nir_lower_tex, &lower_tex_options);

        do {
                progress = false;

                NIR_PASS(progress, nir, nir_lower_var_copies);
                NIR_PASS(progress, nir, nir_lower_vars_to_ssa);

                NIR_PASS(progress, nir, nir_copy_prop);
                NIR_PASS(progress, nir, nir_opt_remove_phis);
                NIR_PASS(progress, nir, nir_opt_dce);
                NIR_PASS(progress, nir, nir_opt_dead_cf);
                NIR_PASS(progress, nir, nir_opt_cse);
                NIR_PASS(progress, nir, nir_opt_peephole_select, 64, false, true);
                NIR_PASS(progress, nir, nir_opt_algebraic);
                NIR_PASS(progress, nir, nir_opt_constant_folding);

                if (lower_flrp != 0) {
                        bool lower_flrp_progress = false;
                        NIR_PASS(lower_flrp_progress,
                                 nir,
                                 nir_lower_flrp,
                                 lower_flrp,
                                 false /* always_precise */,
                                 nir->options->lower_ffma);
                        if (lower_flrp_progress) {
                                NIR_PASS(progress, nir,
                                         nir_opt_constant_folding);
                                progress = true;
                        }

                        /* Nothing should rematerialize any flrps, so we only
                         * need to do this lowering once.
                         */
                        lower_flrp = 0;
                }

                NIR_PASS(progress, nir, nir_opt_undef);
                NIR_PASS(progress, nir, nir_opt_loop_unroll,
                         nir_var_shader_in |
                         nir_var_shader_out |
                         nir_var_function_temp);
        } while (progress);

        NIR_PASS(progress, nir, nir_opt_algebraic_late);

        /* Take us out of SSA */
        NIR_PASS(progress, nir, nir_lower_locals_to_regs);
        NIR_PASS(progress, nir, nir_convert_from_ssa, true);
}

void
bifrost_compile_shader_nir(nir_shader *nir, bifrost_program *program, unsigned product_id)
{
        bi_context *ctx = rzalloc(NULL, bi_context);
        ctx->nir = nir;
        ctx->stage = nir->info.stage;
        ctx->quirks = bifrost_get_quirks(product_id);
        list_inithead(&ctx->blocks);

        /* Lower gl_Position pre-optimisation, but after lowering vars to ssa
         * (so we don't accidentally duplicate the epilogue since mesa/st has
         * messed with our I/O quite a bit already) */

        NIR_PASS_V(nir, nir_lower_vars_to_ssa);

        if (ctx->stage == MESA_SHADER_VERTEX) {
                NIR_PASS_V(nir, nir_lower_viewport_transform);
                NIR_PASS_V(nir, nir_lower_point_size, 1.0, 1024.0);
        }

        NIR_PASS_V(nir, nir_split_var_copies);
        NIR_PASS_V(nir, nir_lower_global_vars_to_local);
        NIR_PASS_V(nir, nir_lower_var_copies);
        NIR_PASS_V(nir, nir_lower_vars_to_ssa);
        NIR_PASS_V(nir, nir_lower_io, nir_var_all, glsl_type_size, 0);
        NIR_PASS_V(nir, nir_lower_ssbo);

        /* We have to lower ALU to scalar ourselves since viewport
         * transformations produce vector ops */
        NIR_PASS_V(nir, nir_lower_alu_to_scalar, NULL, NULL);

        bi_optimize_nir(nir);
        nir_print_shader(nir, stdout);

        nir_foreach_function(func, nir) {
                if (!func->impl)
                        continue;

                emit_cf_list(ctx, &func->impl->body);
                break; /* TODO: Multi-function shaders */
        }

        bi_print_shader(ctx, stdout);

        ralloc_free(ctx);
}