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 namespace aco {
  35 uint64_t debug_flags = 0;
  36
  37 static const struct debug_control aco_debug_options[] = {
  38    {"validateir", DEBUG_VALIDATE},
  39    {"validatera", DEBUG_VALIDATE_RA},
  40    {"perfwarn", DEBUG_PERFWARN},
  41    {NULL, 0}
  42 };
  43
  44 static once_flag init_once_flag = ONCE_FLAG_INIT;
  45
  46 static void init()
  47 {
  48    debug_flags = parse_debug_string(getenv("ACO_DEBUG"), aco_debug_options);
  49
  50    #ifndef NDEBUG
  51    /* enable some flags by default on debug builds */
  52    debug_flags |= aco::DEBUG_VALIDATE;
  53    #endif
  54 }
  55 }
  56
  57 void aco_compile_shader(unsigned shader_count,
  58                         struct nir_shader *const *shaders,
  59                         struct radv_shader_binary **binary,
  60                         struct radv_shader_args *args)
  61 {
  62    call_once(&aco::init_once_flag, aco::init);
  63
  64    ac_shader_config config = {0};
  65    std::unique_ptr<aco::Program> program{new aco::Program};
  66
  67    /* Instruction Selection */
  68    if (args->is_gs_copy_shader)
  69       aco::select_gs_copy_shader(program.get(), shaders[0], &config, args);
  70    else
  71       aco::select_program(program.get(), shader_count, shaders, &config, args);
  72    if (args->options->dump_preoptir) {
  73       std::cerr << "After Instruction Selection:\n";
  74       aco_print_program(program.get(), stderr);
  75    }
  76    aco::validate(program.get(), stderr);
  77
  78    /* Boolean phi lowering */
  79    aco::lower_bool_phis(program.get());
  80    //std::cerr << "After Boolean Phi Lowering:\n";
  81    //aco_print_program(program.get(), stderr);
  82
  83    aco::dominator_tree(program.get());
  84
  85    /* Optimization */
  86    aco::value_numbering(program.get());
  87    aco::optimize(program.get());
  88    aco::validate(program.get(), stderr);
  89
  90    aco::setup_reduce_temp(program.get());
  91    aco::insert_exec_mask(program.get());
  92    aco::validate(program.get(), stderr);
  93
  94    aco::live live_vars = aco::live_var_analysis(program.get(), args->options);
  95    aco::spill(program.get(), live_vars, args->options);
  96
  97    //std::cerr << "Before Schedule:\n";
  98    //aco_print_program(program.get(), stderr);
  99    aco::schedule_program(program.get(), live_vars);
 100
 101    std::string llvm_ir;
 102    if (args->options->record_ir) {
 103       char *data = NULL;
 104       size_t size = 0;
 105       FILE *f = open_memstream(&data, &size);
 106       if (f) {
 107          aco_print_program(program.get(), f);
 108          fputc(0, f);
 109          fclose(f);
 110       }
 111
 112       llvm_ir = std::string(data, data + size);
 113       free(data);
 114    }
 115
 116    /* Register Allocation */
 117    aco::register_allocation(program.get(), live_vars.live_out);
 118    if (args->options->dump_shader) {
 119       std::cerr << "After RA:\n";
 120       aco_print_program(program.get(), stderr);
 121    }
 122
 123    if (aco::validate_ra(program.get(), args->options, stderr)) {
 124       std::cerr << "Program after RA validation failure:\n";
 125       aco_print_program(program.get(), stderr);
 126       abort();
 127    }
 128
 129    aco::ssa_elimination(program.get());
 130    /* Lower to HW Instructions */
 131    aco::lower_to_hw_instr(program.get());
 132    //std::cerr << "After Eliminate Pseudo Instr:\n";
 133    //aco_print_program(program.get(), stderr);
 134
 135    /* Insert Waitcnt */
 136    aco::insert_wait_states(program.get());
 137    aco::insert_NOPs(program.get());
 138
 139    //std::cerr << "After Insert-Waitcnt:\n";
 140    //aco_print_program(program.get(), stderr);
 141
 142    /* Assembly */
 143    std::vector<uint32_t> code;
 144    unsigned exec_size = aco::emit_program(program.get(), code);
 145
 146    bool get_disasm = args->options->dump_shader || args->options->record_ir;
 147
 148    size_t size = llvm_ir.size();
 149
 150    std::string disasm;
 151    if (get_disasm) {
 152       std::ostringstream stream;
 153       aco::print_asm(program.get(), code, exec_size / 4u, stream);
 154       stream << '\0';
 155       disasm = stream.str();
 156       size += disasm.size();
 157    }
 158
 159    size += code.size() * sizeof(uint32_t) + sizeof(radv_shader_binary_legacy);
 160    /* We need to calloc to prevent unintialized data because this will be used
 161     * directly for the disk cache. Uninitialized data can appear because of
 162     * padding in the struct or because legacy_binary->data can be at an offset
 163     * from the start less than sizeof(radv_shader_binary_legacy). */
 164    radv_shader_binary_legacy* legacy_binary = (radv_shader_binary_legacy*) calloc(size, 1);
 165
 166    legacy_binary->base.type = RADV_BINARY_TYPE_LEGACY;
 167    legacy_binary->base.stage = shaders[shader_count-1]->info.stage;
 168    legacy_binary->base.is_gs_copy_shader = args->is_gs_copy_shader;
 169    legacy_binary->base.total_size = size;
 170
 171    memcpy(legacy_binary->data, code.data(), code.size() * sizeof(uint32_t));
 172    legacy_binary->exec_size = exec_size;
 173    legacy_binary->code_size = code.size() * sizeof(uint32_t);
 174
 175    legacy_binary->config = config;
 176    legacy_binary->disasm_size = 0;
 177    legacy_binary->ir_size = llvm_ir.size();
 178
 179    llvm_ir.copy((char*) legacy_binary->data + legacy_binary->code_size, llvm_ir.size());
 180
 181    if (get_disasm) {
 182       disasm.copy((char*) legacy_binary->data + legacy_binary->code_size + llvm_ir.size(), disasm.size());
 183       legacy_binary->disasm_size = disasm.size();
 184    }
 185
 186    *binary = (radv_shader_binary*) legacy_binary;
 187 }