src/amd/compiler/aco_interface.cpp

   1 /*
   2  * Copyright © 2018 Google
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include "aco_interface.h"
  25 #include "aco_ir.h"
  26 #include "vulkan/radv_shader.h"
  27 #include "vulkan/radv_shader_args.h"
  28 #include "c11/threads.h"
  29 #include "util/debug.h"
  30
  31 #include <iostream>
  32 #include <sstream>
  33
  34 static radv_compiler_statistic_info statistic_infos[] = {
  35    [aco::statistic_hash] = {"Hash", "CRC32 hash of code and constant data"},
  36    [aco::statistic_instructions] = {"Instructions", "Instruction count"},
  37    [aco::statistic_copies] = {"Copies", "Copy instructions created for pseudo-instructions"},
  38    [aco::statistic_branches] = {"Branches", "Branch instructions"},
  39    [aco::statistic_cycles] = {"Busy Cycles", "Estimate of busy cycles"},
  40    [aco::statistic_vmem_clauses] = {"VMEM Clause", "Number of VMEM clauses (includes 1-sized clauses)"},
  41    [aco::statistic_smem_clauses] = {"SMEM Clause", "Number of SMEM clauses (includes 1-sized clauses)"},
  42    [aco::statistic_vmem_score] = {"VMEM Score", "Average VMEM def-use distances"},
  43    [aco::statistic_smem_score] = {"SMEM Score", "Average SMEM def-use distances"},
  44    [aco::statistic_sgpr_presched] = {"Pre-Sched SGPRs", "SGPR usage before scheduling"},
  45    [aco::statistic_vgpr_presched] = {"Pre-Sched VGPRs", "VGPR usage before scheduling"},
  46 };
  47
  48 void aco_compile_shader(unsigned shader_count,
  49                         struct nir_shader *const *shaders,
  50                         struct radv_shader_binary **binary,
  51                         struct radv_shader_args *args)
  52 {
  53    aco::init();
  54
  55    ac_shader_config config = {0};
  56    std::unique_ptr<aco::Program> program{new aco::Program};
  57
  58    program->collect_statistics = args->options->record_stats;
  59    if (program->collect_statistics)
  60       memset(program->statistics, 0, sizeof(program->statistics));
  61
  62    /* Instruction Selection */
  63    if (args->is_gs_copy_shader)
  64       aco::select_gs_copy_shader(program.get(), shaders[0], &config, args);
  65    else
  66       aco::select_program(program.get(), shader_count, shaders, &config, args);
  67    if (args->options->dump_preoptir) {
  68       std::cerr << "After Instruction Selection:\n";
  69       aco_print_program(program.get(), stderr);
  70    }
  71
  72    /* Phi lowering */
  73    aco::lower_phis(program.get());
  74    aco::dominator_tree(program.get());
  75    aco::validate(program.get(), stderr);
  76
  77    /* Optimization */
  78    aco::value_numbering(program.get());
  79    aco::optimize(program.get());
  80
  81    /* cleanup and exec mask handling */
  82    aco::setup_reduce_temp(program.get());
  83    aco::insert_exec_mask(program.get());
  84    aco::validate(program.get(), stderr);
  85
  86    /* spilling and scheduling */
  87    aco::live live_vars = aco::live_var_analysis(program.get(), args->options);
  88    aco::spill(program.get(), live_vars, args->options);
  89    if (program->collect_statistics)
  90       aco::collect_presched_stats(program.get());
  91    aco::schedule_program(program.get(), live_vars);
  92    aco::validate(program.get(), stderr);
  93
  94    std::string llvm_ir;
  95    if (args->options->record_ir) {
  96       char *data = NULL;
  97       size_t size = 0;
  98       FILE *f = open_memstream(&data, &size);
  99       if (f) {
 100          aco_print_program(program.get(), f);
 101          fputc(0, f);
 102          fclose(f);
 103       }
 104
 105       llvm_ir = std::string(data, data + size);
 106       free(data);
 107    }
 108
 109    /* Register Allocation */
 110    aco::register_allocation(program.get(), live_vars.live_out);
 111    if (args->options->dump_shader) {
 112       std::cerr << "After RA:\n";
 113       aco_print_program(program.get(), stderr);
 114    }
 115
 116    if (aco::validate_ra(program.get(), args->options, stderr)) {
 117       std::cerr << "Program after RA validation failure:\n";
 118       aco_print_program(program.get(), stderr);
 119       abort();
 120    }
 121
 122    aco::validate(program.get(), stderr);
 123
 124    /* Lower to HW Instructions */
 125    aco::ssa_elimination(program.get());
 126    aco::lower_to_hw_instr(program.get());
 127
 128    /* Insert Waitcnt */
 129    aco::insert_wait_states(program.get());
 130    aco::insert_NOPs(program.get());
 131
 132    if (program->collect_statistics)
 133       aco::collect_preasm_stats(program.get());
 134
 135    /* Assembly */
 136    std::vector<uint32_t> code;
 137    unsigned exec_size = aco::emit_program(program.get(), code);
 138
 139    if (program->collect_statistics)
 140       aco::collect_postasm_stats(program.get(), code);
 141
 142    bool get_disasm = args->options->dump_shader || args->options->record_ir;
 143
 144    size_t size = llvm_ir.size();
 145
 146    std::string disasm;
 147    if (get_disasm) {
 148       std::ostringstream stream;
 149       aco::print_asm(program.get(), code, exec_size / 4u, stream);
 150       stream << '\0';
 151       disasm = stream.str();
 152       size += disasm.size();
 153    }
 154
 155    size_t stats_size = 0;
 156    if (program->collect_statistics)
 157       stats_size = sizeof(radv_compiler_statistics) + aco::num_statistics * sizeof(uint32_t);
 158    size += stats_size;
 159
 160    size += code.size() * sizeof(uint32_t) + sizeof(radv_shader_binary_legacy);
 161    /* We need to calloc to prevent unintialized data because this will be used
 162     * directly for the disk cache. Uninitialized data can appear because of
 163     * padding in the struct or because legacy_binary->data can be at an offset
 164     * from the start less than sizeof(radv_shader_binary_legacy). */
 165    radv_shader_binary_legacy* legacy_binary = (radv_shader_binary_legacy*) calloc(size, 1);
 166
 167    legacy_binary->base.type = RADV_BINARY_TYPE_LEGACY;
 168    legacy_binary->base.stage = shaders[shader_count-1]->info.stage;
 169    legacy_binary->base.is_gs_copy_shader = args->is_gs_copy_shader;
 170    legacy_binary->base.total_size = size;
 171
 172    if (program->collect_statistics) {
 173       radv_compiler_statistics *statistics = (radv_compiler_statistics *)legacy_binary->data;
 174       statistics->count = aco::num_statistics;
 175       statistics->infos = statistic_infos;
 176       memcpy(statistics->values, program->statistics, aco::num_statistics * sizeof(uint32_t));
 177    }
 178    legacy_binary->stats_size = stats_size;
 179
 180    memcpy(legacy_binary->data + legacy_binary->stats_size, code.data(), code.size() * sizeof(uint32_t));
 181    legacy_binary->exec_size = exec_size;
 182    legacy_binary->code_size = code.size() * sizeof(uint32_t);
 183
 184    legacy_binary->config = config;
 185    legacy_binary->disasm_size = 0;
 186    legacy_binary->ir_size = llvm_ir.size();
 187
 188    llvm_ir.copy((char*) legacy_binary->data + legacy_binary->stats_size + legacy_binary->code_size, llvm_ir.size());
 189
 190    if (get_disasm) {
 191       disasm.copy((char*) legacy_binary->data + legacy_binary->stats_size + legacy_binary->code_size + llvm_ir.size(), disasm.size());
 192       legacy_binary->disasm_size = disasm.size();
 193    }
 194
 195    *binary = (radv_shader_binary*) legacy_binary;
 196 }