src/amd/compiler/aco_ssa_elimination.cpp

   1 /*
   2  * Copyright © 2018 Valve Corporation
   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
  25
  26 #include "aco_ir.h"
  27
  28 #include <map>
  29
  30 namespace aco {
  31 namespace {
  32
  33 /* map: block-id -> pair (dest, src) to store phi information */
  34 typedef std::map<uint32_t, std::vector<std::pair<Definition, Operand>>> phi_info;
  35
  36 struct ssa_elimination_ctx {
  37    phi_info logical_phi_info;
  38    phi_info linear_phi_info;
  39    std::vector<bool> empty_blocks;
  40    Program* program;
  41
  42    ssa_elimination_ctx(Program* program) : empty_blocks(program->blocks.size(), true), program(program) {}
  43 };
  44
  45 void collect_phi_info(ssa_elimination_ctx& ctx)
  46 {
  47    for (Block& block : ctx.program->blocks) {
  48       for (aco_ptr<Instruction>& phi : block.instructions) {
  49          if (phi->opcode != aco_opcode::p_phi && phi->opcode != aco_opcode::p_linear_phi)
  50             break;
  51
  52          for (unsigned i = 0; i < phi->operands.size(); i++) {
  53             if (phi->operands[i].isUndefined())
  54                continue;
  55             if (phi->operands[i].isTemp() && phi->operands[i].physReg() == phi->definitions[0].physReg())
  56                continue;
  57
  58             std::vector<unsigned>& preds = phi->opcode == aco_opcode::p_phi ? block.logical_preds : block.linear_preds;
  59             phi_info& info = phi->opcode == aco_opcode::p_phi ? ctx.logical_phi_info : ctx.linear_phi_info;
  60             const auto result = info.emplace(preds[i], std::vector<std::pair<Definition, Operand>>());
  61             result.first->second.emplace_back(phi->definitions[0], phi->operands[i]);
  62             ctx.empty_blocks[preds[i]] = false;
  63          }
  64       }
  65    }
  66 }
  67
  68 void insert_parallelcopies(ssa_elimination_ctx& ctx)
  69 {
  70    /* insert the parallelcopies from logical phis before p_logical_end */
  71    for (auto&& entry : ctx.logical_phi_info) {
  72       Block& block = ctx.program->blocks[entry.first];
  73       unsigned idx = block.instructions.size() - 1;
  74       while (block.instructions[idx]->opcode != aco_opcode::p_logical_end) {
  75          assert(idx > 0);
  76          idx--;
  77       }
  78
  79       std::vector<aco_ptr<Instruction>>::iterator it = std::next(block.instructions.begin(), idx);
  80       aco_ptr<Pseudo_instruction> pc{create_instruction<Pseudo_instruction>(aco_opcode::p_parallelcopy, Format::PSEUDO, entry.second.size(), entry.second.size())};
  81       unsigned i = 0;
  82       for (std::pair<Definition, Operand>& pair : entry.second)
  83       {
  84          pc->definitions[i] = pair.first;
  85          pc->operands[i] = pair.second;
  86          i++;
  87       }
  88       /* this shouldn't be needed since we're only copying vgprs */
  89       pc->tmp_in_scc = false;
  90       block.instructions.insert(it, std::move(pc));
  91    }
  92
  93    /* insert parallelcopies for the linear phis at the end of blocks just before the branch */
  94    for (auto&& entry : ctx.linear_phi_info) {
  95       Block& block = ctx.program->blocks[entry.first];
  96       std::vector<aco_ptr<Instruction>>::iterator it = block.instructions.end();
  97       --it;
  98       assert((*it)->format == Format::PSEUDO_BRANCH);
  99       aco_ptr<Pseudo_instruction> pc{create_instruction<Pseudo_instruction>(aco_opcode::p_parallelcopy, Format::PSEUDO, entry.second.size(), entry.second.size())};
 100       unsigned i = 0;
 101       for (std::pair<Definition, Operand>& pair : entry.second)
 102       {
 103          pc->definitions[i] = pair.first;
 104          pc->operands[i] = pair.second;
 105          i++;
 106       }
 107       pc->tmp_in_scc = block.scc_live_out;
 108       pc->scratch_sgpr = block.scratch_sgpr;
 109       block.instructions.insert(it, std::move(pc));
 110    }
 111 }
 112
 113
 114 void try_remove_merge_block(ssa_elimination_ctx& ctx, Block* block)
 115 {
 116    /* check if the successor is another merge block which restores exec */
 117    // TODO: divergent loops also restore exec
 118    if (block->linear_succs.size() != 1 ||
 119        !(ctx.program->blocks[block->linear_succs[0]].kind & block_kind_merge))
 120       return;
 121
 122    /* check if this block is empty and the exec mask is not needed */
 123    for (aco_ptr<Instruction>& instr : block->instructions) {
 124       if (instr->opcode == aco_opcode::p_parallelcopy) {
 125          if (instr->definitions[0].physReg() == exec)
 126             continue;
 127          else
 128             return;
 129       }
 130
 131       if (instr->opcode != aco_opcode::p_linear_phi &&
 132           instr->opcode != aco_opcode::p_phi &&
 133           instr->opcode != aco_opcode::p_logical_start &&
 134           instr->opcode != aco_opcode::p_logical_end &&
 135           instr->opcode != aco_opcode::p_branch)
 136          return;
 137    }
 138
 139    /* keep the branch instruction and remove the rest */
 140    aco_ptr<Instruction> branch = std::move(block->instructions.back());
 141    block->instructions.clear();
 142    block->instructions.emplace_back(std::move(branch));
 143 }
 144
 145 void try_remove_invert_block(ssa_elimination_ctx& ctx, Block* block)
 146 {
 147    assert(block->linear_succs.size() == 2);
 148    if (block->linear_succs[0] != block->linear_succs[1])
 149       return;
 150
 151    /* check if we can remove this block */
 152    for (aco_ptr<Instruction>& instr : block->instructions) {
 153       if (instr->opcode != aco_opcode::p_linear_phi &&
 154           instr->opcode != aco_opcode::p_phi &&
 155           instr->opcode != aco_opcode::s_andn2_b64 &&
 156           instr->opcode != aco_opcode::p_branch)
 157          return;
 158    }
 159
 160    unsigned succ_idx = block->linear_succs[0];
 161    assert(block->linear_preds.size() == 2);
 162    for (unsigned i = 0; i < 2; i++) {
 163       Block *pred = &ctx.program->blocks[block->linear_preds[i]];
 164       pred->linear_succs[0] = succ_idx;
 165       ctx.program->blocks[succ_idx].linear_preds[i] = pred->index;
 166
 167       Pseudo_branch_instruction *branch = static_cast<Pseudo_branch_instruction*>(pred->instructions.back().get());
 168       assert(branch->format == Format::PSEUDO_BRANCH);
 169       branch->target[0] = succ_idx;
 170       branch->target[1] = succ_idx;
 171    }
 172
 173    block->instructions.clear();
 174    block->linear_preds.clear();
 175    block->linear_succs.clear();
 176 }
 177
 178 void try_remove_simple_block(ssa_elimination_ctx& ctx, Block* block)
 179 {
 180    for (aco_ptr<Instruction>& instr : block->instructions) {
 181       if (instr->opcode != aco_opcode::p_logical_start &&
 182           instr->opcode != aco_opcode::p_logical_end &&
 183           instr->opcode != aco_opcode::p_branch)
 184          return;
 185    }
 186
 187    Block& pred = ctx.program->blocks[block->linear_preds[0]];
 188    Block& succ = ctx.program->blocks[block->linear_succs[0]];
 189    Pseudo_branch_instruction* branch = static_cast<Pseudo_branch_instruction*>(pred.instructions.back().get());
 190    if (branch->opcode == aco_opcode::p_branch) {
 191       branch->target[0] = succ.index;
 192       branch->target[1] = succ.index;
 193    } else if (branch->target[0] == block->index) {
 194       branch->target[0] = succ.index;
 195    } else if (branch->target[0] == succ.index) {
 196       assert(branch->target[1] == block->index);
 197       branch->target[1] = succ.index;
 198       branch->opcode = aco_opcode::p_branch;
 199    } else if (branch->target[1] == block->index) {
 200       /* check if there is a fall-through path from block to succ */
 201       bool falls_through = true;
 202       for (unsigned j = block->index + 1; falls_through && j < succ.index; j++) {
 203          assert(ctx.program->blocks[j].index == j);
 204          if (!ctx.program->blocks[j].instructions.empty())
 205             falls_through = false;
 206       }
 207       if (falls_through) {
 208          branch->target[1] = succ.index;
 209       } else {
 210          /* check if there is a fall-through path for the alternative target */
 211          for (unsigned j = block->index + 1; j < branch->target[0]; j++) {
 212             if (!ctx.program->blocks[j].instructions.empty())
 213                return;
 214          }
 215
 216          /* This is a (uniform) break or continue block. The branch condition has to be inverted. */
 217          if (branch->opcode == aco_opcode::p_cbranch_z)
 218             branch->opcode = aco_opcode::p_cbranch_nz;
 219          else if (branch->opcode == aco_opcode::p_cbranch_nz)
 220             branch->opcode = aco_opcode::p_cbranch_z;
 221          else
 222             assert(false);
 223          /* also invert the linear successors */
 224          pred.linear_succs[0] = pred.linear_succs[1];
 225          pred.linear_succs[1] = succ.index;
 226          branch->target[1] = branch->target[0];
 227          branch->target[0] = succ.index;
 228       }
 229    } else {
 230       assert(false);
 231    }
 232
 233    if (branch->target[0] == branch->target[1])
 234       branch->opcode = aco_opcode::p_branch;
 235
 236    for (unsigned i = 0; i < pred.linear_succs.size(); i++)
 237       if (pred.linear_succs[i] == block->index)
 238          pred.linear_succs[i] = succ.index;
 239
 240    for (unsigned i = 0; i < succ.linear_preds.size(); i++)
 241       if (succ.linear_preds[i] == block->index)
 242          succ.linear_preds[i] = pred.index;
 243
 244    block->instructions.clear();
 245    block->linear_preds.clear();
 246    block->linear_succs.clear();
 247 }
 248
 249 void jump_threading(ssa_elimination_ctx& ctx)
 250 {
 251    for (int i = ctx.program->blocks.size() - 1; i >= 0; i--) {
 252       Block* block = &ctx.program->blocks[i];
 253
 254       if (!ctx.empty_blocks[i])
 255          continue;
 256
 257       if (block->kind & block_kind_invert) {
 258          try_remove_invert_block(ctx, block);
 259          continue;
 260       }
 261
 262       if (block->linear_succs.size() > 1)
 263          continue;
 264
 265       if (block->kind & block_kind_merge ||
 266           block->kind & block_kind_loop_exit)
 267          try_remove_merge_block(ctx, block);
 268
 269       if (block->linear_preds.size() == 1)
 270          try_remove_simple_block(ctx, block);
 271    }
 272 }
 273
 274 } /* end namespace */
 275
 276
 277 void ssa_elimination(Program* program)
 278 {
 279    ssa_elimination_ctx ctx(program);
 280
 281    /* Collect information about every phi-instruction */
 282    collect_phi_info(ctx);
 283
 284    /* eliminate empty blocks */
 285    jump_threading(ctx);
 286
 287    /* insert parallelcopies from SSA elimination */
 288    insert_parallelcopies(ctx);
 289
 290 }
 291 }