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    aco::select_program(program.get(), shader_count, shaders, &config,
  69                        args->shader_info, args->options);
  70    if (args->options->dump_preoptir) {
  71       std::cerr << "After Instruction Selection:\n";
  72       aco_print_program(program.get(), stderr);
  73    }
  74    aco::validate(program.get(), stderr);
  75
  76    /* Boolean phi lowering */
  77    aco::lower_bool_phis(program.get());
  78    //std::cerr << "After Boolean Phi Lowering:\n";
  79    //aco_print_program(program.get(), stderr);
  80
  81    aco::dominator_tree(program.get());
  82
  83    /* Optimization */
  84    aco::value_numbering(program.get());
  85    aco::optimize(program.get());
  86    aco::validate(program.get(), stderr);
  87
  88    aco::setup_reduce_temp(program.get());
  89    aco::insert_exec_mask(program.get());
  90    aco::validate(program.get(), stderr);
  91
  92    aco::live live_vars = aco::live_var_analysis(program.get(), args->options);
  93    aco::spill(program.get(), live_vars, args->options);
  94
  95    //std::cerr << "Before Schedule:\n";
  96    //aco_print_program(program.get(), stderr);
  97    aco::schedule_program(program.get(), live_vars);
  98
  99    std::string llvm_ir;
 100    if (args->options->record_ir) {
 101       char *data = NULL;
 102       size_t size = 0;
 103       FILE *f = open_memstream(&data, &size);
 104       if (f) {
 105          aco_print_program(program.get(), f);
 106          fputc(0, f);
 107          fclose(f);
 108       }
 109
 110       llvm_ir = std::string(data, data + size);
 111       free(data);
 112    }
 113
 114    /* Register Allocation */
 115    aco::register_allocation(program.get(), live_vars.live_out);
 116    if (args->options->dump_shader) {
 117       std::cerr << "After RA:\n";
 118       aco_print_program(program.get(), stderr);
 119    }
 120
 121    if (aco::validate_ra(program.get(), args->options, stderr)) {
 122       std::cerr << "Program after RA validation failure:\n";
 123       aco_print_program(program.get(), stderr);
 124       abort();
 125    }
 126
 127    aco::ssa_elimination(program.get());
 128    /* Lower to HW Instructions */
 129    aco::lower_to_hw_instr(program.get());
 130    //std::cerr << "After Eliminate Pseudo Instr:\n";
 131    //aco_print_program(program.get(), stderr);
 132
 133    /* Insert Waitcnt */
 134    aco::insert_wait_states(program.get());
 135    aco::insert_NOPs(program.get());
 136
 137    //std::cerr << "After Insert-Waitcnt:\n";
 138    //aco_print_program(program.get(), stderr);
 139
 140    /* Assembly */
 141    std::vector<uint32_t> code;
 142    unsigned exec_size = aco::emit_program(program.get(), code);
 143
 144    bool get_disasm = args->options->dump_shader || args->options->record_ir;
 145
 146    size_t size = llvm_ir.size();
 147
 148    std::string disasm;
 149    if (get_disasm) {
 150       std::ostringstream stream;
 151       aco::print_asm(program.get(), code, exec_size / 4u, stream);
 152       stream << '\0';
 153       disasm = stream.str();
 154       size += disasm.size();
 155    }
 156
 157    size += code.size() * sizeof(uint32_t) + sizeof(radv_shader_binary_legacy);
 158    radv_shader_binary_legacy* legacy_binary = (radv_shader_binary_legacy*) malloc(size);
 159
 160    legacy_binary->base.type = RADV_BINARY_TYPE_LEGACY;
 161    legacy_binary->base.stage = shaders[shader_count-1]->info.stage;
 162    legacy_binary->base.is_gs_copy_shader = false;
 163    legacy_binary->base.total_size = size;
 164
 165    memcpy(legacy_binary->data, code.data(), code.size() * sizeof(uint32_t));
 166    legacy_binary->exec_size = exec_size;
 167    legacy_binary->code_size = code.size() * sizeof(uint32_t);
 168
 169    legacy_binary->config = config;
 170    legacy_binary->disasm_size = 0;
 171    legacy_binary->ir_size = llvm_ir.size();
 172
 173    llvm_ir.copy((char*) legacy_binary->data + legacy_binary->code_size, llvm_ir.size());
 174
 175    if (get_disasm) {
 176       disasm.copy((char*) legacy_binary->data + legacy_binary->code_size + llvm_ir.size(), disasm.size());
 177       legacy_binary->disasm_size = disasm.size();
 178    }
 179
 180    *binary = (radv_shader_binary*) legacy_binary;
 181 }