mesa/src/amd/compiler/aco_live_var_analysis.cpp

/*
 * Copyright © 2018 Valve Corporation
 * Copyright © 2018 Google
 *
 * SPDX-License-Identifier: MIT
 */

#include "aco_ir.h"

namespace aco {

RegisterDemand
get_live_changes(Instruction* instr)
{
   RegisterDemand changes;
   for (const Definition& def : instr->definitions) {
      if (!def.isTemp() || def.isKill())
         continue;
      changes += def.getTemp();
   }

   for (const Operand& op : instr->operands) {
      if (!op.isTemp() || !op.isFirstKill())
         continue;
      changes -= op.getTemp();
   }

   return changes;
}

RegisterDemand
get_additional_operand_demand(Instruction* instr)
{
   RegisterDemand additional_demand;
   int op_idx = get_op_fixed_to_def(instr);
   if (op_idx != -1 && !instr->operands[op_idx].isKill())
      additional_demand += instr->definitions[0].getTemp();

   return additional_demand;
}

RegisterDemand
get_temp_registers(Instruction* instr)
{
   RegisterDemand demand_before;
   RegisterDemand demand_after;

   for (Definition def : instr->definitions) {
      if (def.isKill())
         demand_after += def.getTemp();
      else if (def.isTemp())
         demand_before -= def.getTemp();
   }

   for (Operand op : instr->operands) {
      if (op.isFirstKill()) {
         demand_before += op.getTemp();
         if (op.isLateKill())
            demand_after += op.getTemp();
      }
   }

   demand_before += get_additional_operand_demand(instr);
   demand_after.update(demand_before);
   return demand_after;
}

namespace {

struct live_ctx {
   monotonic_buffer_resource m;
   Program* program;
   int32_t worklist;
   uint32_t handled_once;
};

bool
instr_needs_vcc(Instruction* instr)
{
   if (instr->isVOPC())
      return true;
   if (instr->isVOP2() && !instr->isVOP3()) {
      if (instr->operands.size() == 3 && instr->operands[2].isTemp() &&
          instr->operands[2].regClass().type() == RegType::sgpr)
         return true;
      if (instr->definitions.size() == 2)
         return true;
   }
   return false;
}

IDSet
compute_live_out(live_ctx& ctx, Block* block)
{
   IDSet live(ctx.m);

   if (block->logical_succs.empty()) {
      /* Linear blocks:
       * Directly insert the successor if it is a linear block as well.
       */
      for (unsigned succ : block->linear_succs) {
         if (ctx.program->blocks[succ].logical_preds.empty()) {
            live.insert(ctx.program->live.live_in[succ]);
         } else {
            for (unsigned t : ctx.program->live.live_in[succ]) {
               if (ctx.program->temp_rc[t].is_linear())
                  live.insert(t);
            }
         }
      }
   } else {
      /* Logical blocks:
       * Linear successors are either linear blocks or logical targets.
       */
      live = IDSet(ctx.program->live.live_in[block->linear_succs[0]], ctx.m);
      if (block->linear_succs.size() == 2)
         live.insert(ctx.program->live.live_in[block->linear_succs[1]]);

      /* At most one logical target needs a separate insertion. */
      if (block->logical_succs.back() != block->linear_succs.back()) {
         for (unsigned t : ctx.program->live.live_in[block->logical_succs.back()]) {
            if (!ctx.program->temp_rc[t].is_linear())
               live.insert(t);
         }
      } else {
         assert(block->logical_succs[0] == block->linear_succs[0]);
      }
   }

   /* Handle phi operands */
   if (block->linear_succs.size() == 1 && block->linear_succs[0] >= ctx.handled_once) {
      Block& succ = ctx.program->blocks[block->linear_succs[0]];
      auto it = std::find(succ.linear_preds.begin(), succ.linear_preds.end(), block->index);
      unsigned op_idx = std::distance(succ.linear_preds.begin(), it);
      for (aco_ptr<Instruction>& phi : succ.instructions) {
         if (!is_phi(phi))
            break;
         if (phi->opcode == aco_opcode::p_phi || phi->definitions[0].isKill())
            continue;
         if (phi->operands[op_idx].isTemp())
            live.insert(phi->operands[op_idx].tempId());
      }
   }
   if (block->logical_succs.size() == 1 && block->logical_succs[0] >= ctx.handled_once) {
      Block& succ = ctx.program->blocks[block->logical_succs[0]];
      auto it = std::find(succ.logical_preds.begin(), succ.logical_preds.end(), block->index);
      unsigned op_idx = std::distance(succ.logical_preds.begin(), it);
      for (aco_ptr<Instruction>& phi : succ.instructions) {
         if (!is_phi(phi))
            break;
         if (phi->opcode == aco_opcode::p_linear_phi || phi->definitions[0].isKill())
            continue;
         if (phi->operands[op_idx].isTemp())
            live.insert(phi->operands[op_idx].tempId());
      }
   }

   return live;
}

void
process_live_temps_per_block(live_ctx& ctx, Block* block)
{
   RegisterDemand new_demand;
   block->register_demand = RegisterDemand();
   IDSet live = compute_live_out(ctx, block);

   /* initialize register demand */
   for (unsigned t : live)
      new_demand += Temp(t, ctx.program->temp_rc[t]);

   /* traverse the instructions backwards */
   int idx;
   for (idx = block->instructions.size() - 1; idx >= 0; idx--) {
      Instruction* insn = block->instructions[idx].get();
      if (is_phi(insn))
         break;

      ctx.program->needs_vcc |= instr_needs_vcc(insn);
      insn->register_demand = RegisterDemand(new_demand.vgpr, new_demand.sgpr);

      bool has_vgpr_def = false;

      /* KILL */
      for (Definition& definition : insn->definitions) {
         has_vgpr_def |= definition.regClass().type() == RegType::vgpr &&
                         !definition.regClass().is_linear_vgpr();

         if (!definition.isTemp()) {
            continue;
         }
         if (definition.isFixed() && definition.physReg() == vcc)
            ctx.program->needs_vcc = true;

         const Temp temp = definition.getTemp();
         const size_t n = live.erase(temp.id());

         if (n) {
            new_demand -= temp;
            definition.setKill(false);
         } else {
            insn->register_demand += temp;
            definition.setKill(true);
         }
      }

      if (ctx.program->gfx_level >= GFX10 && insn->isVALU() &&
          insn->definitions.back().regClass() == s2) {
         /* RDNA2 ISA doc, 6.2.4. Wave64 Destination Restrictions:
          * The first pass of a wave64 VALU instruction may not overwrite a scalar value used by
          * the second half.
          */
         bool carry_in = insn->opcode == aco_opcode::v_addc_co_u32 ||
                         insn->opcode == aco_opcode::v_subb_co_u32 ||
                         insn->opcode == aco_opcode::v_subbrev_co_u32;
         for (unsigned op_idx = 0; op_idx < (carry_in ? 2 : insn->operands.size()); op_idx++) {
            if (insn->operands[op_idx].isOfType(RegType::sgpr))
               insn->operands[op_idx].setLateKill(true);
         }
      } else if (insn->opcode == aco_opcode::p_bpermute_readlane ||
                 insn->opcode == aco_opcode::p_bpermute_permlane ||
                 insn->opcode == aco_opcode::p_bpermute_shared_vgpr ||
                 insn->opcode == aco_opcode::p_dual_src_export_gfx11 ||
                 insn->opcode == aco_opcode::v_mqsad_u32_u8) {
         for (Operand& op : insn->operands)
            op.setLateKill(true);
      } else if (insn->opcode == aco_opcode::p_interp_gfx11) {
         insn->operands.back().setLateKill(true); /* we don't want the bld.lm def to use m0 */
         if (insn->operands.size() == 7)
            insn->operands[5].setLateKill(true); /* we re-use the destination reg in the middle */
      } else if (insn->opcode == aco_opcode::v_interp_p1_f32 && ctx.program->dev.has_16bank_lds) {
         insn->operands[0].setLateKill(true);
      } else if (insn->opcode == aco_opcode::p_init_scratch) {
         insn->operands.back().setLateKill(true);
      } else if (instr_info.classes[(int)insn->opcode] == instr_class::wmma) {
         insn->operands[0].setLateKill(true);
         insn->operands[1].setLateKill(true);
      }

      /* Check if a definition clobbers some operand */
      int op_idx = get_op_fixed_to_def(insn);
      if (op_idx != -1)
         insn->operands[op_idx].setClobbered(true);

      /* we need to do this in a separate loop because the next one can
       * setKill() for several operands at once and we don't want to
       * overwrite that in a later iteration */
      for (Operand& op : insn->operands) {
         op.setKill(false);
         /* Linear vgprs must be late kill: this is to ensure linear VGPR operands and
          * normal VGPR definitions don't try to use the same register, which is problematic
          * because of assignment restrictions.
          */
         if (op.hasRegClass() && op.regClass().is_linear_vgpr() && !op.isUndefined() &&
             has_vgpr_def)
            op.setLateKill(true);
      }

      /* GEN */
      for (unsigned i = 0; i < insn->operands.size(); ++i) {
         Operand& operand = insn->operands[i];
         if (!operand.isTemp())
            continue;
         if (operand.isFixed() && ctx.program->progress < CompilationProgress::after_ra) {
            ctx.program->needs_vcc |= operand.physReg() == vcc;
            /* Check if this operand gets overwritten by a precolored definition. */
            if (std::any_of(insn->definitions.begin(), insn->definitions.end(),
                            [=](Definition def)
                            {
                               return def.isFixed() &&
                                      def.physReg() + def.size() > operand.physReg() &&
                                      operand.physReg() + operand.size() > def.physReg();
                            }))
               operand.setClobbered(true);
         }
         const Temp temp = operand.getTemp();

         if (operand.isKill())
            continue;

         if (live.insert(temp.id()).second) {
            operand.setFirstKill(true);
            for (unsigned j = i + 1; j < insn->operands.size(); ++j) {
               if (insn->operands[j].isTemp() && insn->operands[j].getTemp() == temp)
                  insn->operands[j].setKill(true);
            }
            if (operand.isLateKill())
               insn->register_demand += temp;
            new_demand += temp;
         }
      }

      RegisterDemand before_instr = new_demand + get_additional_operand_demand(insn);
      insn->register_demand.update(before_instr);
      block->register_demand.update(insn->register_demand);
   }

   /* handle phi definitions */
   for (int phi_idx = 0; phi_idx <= idx; phi_idx++) {
      Instruction* insn = block->instructions[phi_idx].get();
      insn->register_demand = new_demand;

      assert(is_phi(insn) && insn->definitions.size() == 1);
      if (!insn->definitions[0].isTemp()) {
         assert(insn->definitions[0].isFixed() && insn->definitions[0].physReg() == exec);
         continue;
      }
      Definition& definition = insn->definitions[0];
      if (definition.isFixed() && definition.physReg() == vcc)
         ctx.program->needs_vcc = true;
      const size_t n = live.erase(definition.tempId());
      if (n && (definition.isKill() || ctx.handled_once > block->index)) {
         Block::edge_vec& preds =
            insn->opcode == aco_opcode::p_phi ? block->logical_preds : block->linear_preds;
         for (unsigned i = 0; i < preds.size(); i++) {
            if (insn->operands[i].isTemp())
               ctx.worklist = std::max<int>(ctx.worklist, preds[i]);
         }
      }
      definition.setKill(!n);
   }

   /* handle phi operands */
   for (int phi_idx = 0; phi_idx <= idx; phi_idx++) {
      Instruction* insn = block->instructions[phi_idx].get();
      assert(is_phi(insn));
      /* Ignore dead phis. */
      if (insn->definitions[0].isKill())
         continue;
      for (Operand& operand : insn->operands) {
         if (!operand.isTemp())
            continue;
         if (operand.isFixed() && operand.physReg() == vcc)
            ctx.program->needs_vcc = true;

         /* set if the operand is killed by this (or another) phi instruction */
         operand.setKill(!live.count(operand.tempId()));
      }
   }

   if (ctx.program->live.live_in[block->index].insert(live)) {
      if (block->linear_preds.size()) {
         assert(block->logical_preds.empty() ||
                block->logical_preds.back() <= block->linear_preds.back());
         ctx.worklist = std::max<int>(ctx.worklist, block->linear_preds.back());
      } else {
         ASSERTED bool is_valid = validate_ir(ctx.program);
         assert(!is_valid);
      }
   }

   block->live_in_demand = new_demand;
   block->live_in_demand.sgpr += 2; /* Add 2 SGPRs for potential long-jumps. */
   block->register_demand.update(block->live_in_demand);
   ctx.program->max_reg_demand.update(block->register_demand);
   ctx.handled_once = std::min(ctx.handled_once, block->index);

   assert(!block->linear_preds.empty() || (new_demand == RegisterDemand() && live.empty()));
}

unsigned
calc_waves_per_workgroup(Program* program)
{
   /* When workgroup size is not known, just go with wave_size */
   unsigned workgroup_size =
      program->workgroup_size == UINT_MAX ? program->wave_size : program->workgroup_size;

   return align(workgroup_size, program->wave_size) / program->wave_size;
}
} /* end namespace */

bool
uses_scratch(Program* program)
{
   /* RT uses scratch but we don't yet know how much. */
   return program->config->scratch_bytes_per_wave || program->stage == raytracing_cs;
}

uint16_t
get_extra_sgprs(Program* program)
{
   /* We don't use this register on GFX6-8 and it's removed on GFX10+. */
   bool needs_flat_scr = uses_scratch(program) && program->gfx_level == GFX9;

   if (program->gfx_level >= GFX10) {
      assert(!program->dev.xnack_enabled);
      return 0;
   } else if (program->gfx_level >= GFX8) {
      if (needs_flat_scr)
         return 6;
      else if (program->dev.xnack_enabled)
         return 4;
      else if (program->needs_vcc)
         return 2;
      else
         return 0;
   } else {
      assert(!program->dev.xnack_enabled);
      if (needs_flat_scr)
         return 4;
      else if (program->needs_vcc)
         return 2;
      else
         return 0;
   }
}

uint16_t
get_sgpr_alloc(Program* program, uint16_t addressable_sgprs)
{
   uint16_t sgprs = addressable_sgprs + get_extra_sgprs(program);
   uint16_t granule = program->dev.sgpr_alloc_granule;
   return ALIGN_NPOT(std::max(sgprs, granule), granule);
}

uint16_t
get_vgpr_alloc(Program* program, uint16_t addressable_vgprs)
{
   assert(addressable_vgprs <= program->dev.vgpr_limit);
   uint16_t granule = program->dev.vgpr_alloc_granule;
   return ALIGN_NPOT(std::max(addressable_vgprs, granule), granule);
}

unsigned
round_down(unsigned a, unsigned b)
{
   return a - (a % b);
}

uint16_t
get_addr_sgpr_from_waves(Program* program, uint16_t waves)
{
   /* it's not possible to allocate more than 128 SGPRs */
   uint16_t sgprs = std::min(program->dev.physical_sgprs / waves, 128);
   sgprs = round_down(sgprs, program->dev.sgpr_alloc_granule);
   sgprs -= get_extra_sgprs(program);
   return std::min(sgprs, program->dev.sgpr_limit);
}

uint16_t
get_addr_vgpr_from_waves(Program* program, uint16_t waves)
{
   uint16_t vgprs = program->dev.physical_vgprs / waves;
   vgprs = vgprs / program->dev.vgpr_alloc_granule * program->dev.vgpr_alloc_granule;
   vgprs -= program->config->num_shared_vgprs / 2;
   return std::min(vgprs, program->dev.vgpr_limit);
}

void
calc_min_waves(Program* program)
{
   unsigned waves_per_workgroup = calc_waves_per_workgroup(program);
   unsigned simd_per_cu_wgp = program->dev.simd_per_cu * (program->wgp_mode ? 2 : 1);
   program->min_waves = DIV_ROUND_UP(waves_per_workgroup, simd_per_cu_wgp);
}

uint16_t
max_suitable_waves(Program* program, uint16_t waves)
{
   unsigned num_simd = program->dev.simd_per_cu * (program->wgp_mode ? 2 : 1);
   unsigned waves_per_workgroup = calc_waves_per_workgroup(program);
   unsigned num_workgroups = waves * num_simd / waves_per_workgroup;

   /* Adjust #workgroups for LDS */
   unsigned lds_per_workgroup = align(program->config->lds_size * program->dev.lds_encoding_granule,
                                      program->dev.lds_alloc_granule);

   if (program->stage == fragment_fs) {
      /* PS inputs are moved from PC (parameter cache) to LDS before PS waves are launched.
       * Each PS input occupies 3x vec4 of LDS space. See Figure 10.3 in GCN3 ISA manual.
       * These limit occupancy the same way as other stages' LDS usage does.
       */
      unsigned lds_bytes_per_interp = 3 * 16;
      unsigned lds_param_bytes = lds_bytes_per_interp * program->info.ps.num_interp;
      lds_per_workgroup += align(lds_param_bytes, program->dev.lds_alloc_granule);
   }
   unsigned lds_limit = program->wgp_mode ? program->dev.lds_limit * 2 : program->dev.lds_limit;
   if (lds_per_workgroup)
      num_workgroups = std::min(num_workgroups, lds_limit / lds_per_workgroup);

   /* Hardware limitation */
   if (waves_per_workgroup > 1)
      num_workgroups = std::min(num_workgroups, program->wgp_mode ? 32u : 16u);

   /* Adjust #waves for workgroup multiples:
    * In cases like waves_per_workgroup=3 or lds=65536 and
    * waves_per_workgroup=1, we want the maximum possible number of waves per
    * SIMD and not the minimum. so DIV_ROUND_UP is used
    */
   unsigned workgroup_waves = num_workgroups * waves_per_workgroup;
   return DIV_ROUND_UP(workgroup_waves, num_simd);
}

void
update_vgpr_sgpr_demand(Program* program, const RegisterDemand new_demand)
{
   assert(program->min_waves >= 1);
   uint16_t sgpr_limit = get_addr_sgpr_from_waves(program, program->min_waves);
   uint16_t vgpr_limit = get_addr_vgpr_from_waves(program, program->min_waves);

   /* this won't compile, register pressure reduction necessary */
   if (new_demand.vgpr > vgpr_limit || new_demand.sgpr > sgpr_limit) {
      program->num_waves = 0;
      program->max_reg_demand = new_demand;
   } else {
      program->num_waves = program->dev.physical_sgprs / get_sgpr_alloc(program, new_demand.sgpr);
      uint16_t vgpr_demand =
         get_vgpr_alloc(program, new_demand.vgpr) + program->config->num_shared_vgprs / 2;
      program->num_waves =
         std::min<uint16_t>(program->num_waves, program->dev.physical_vgprs / vgpr_demand);
      program->num_waves = std::min(program->num_waves, program->dev.max_waves_per_simd);

      /* Adjust for LDS and workgroup multiples and calculate max_reg_demand */
      program->num_waves = max_suitable_waves(program, program->num_waves);
      program->max_reg_demand.vgpr = get_addr_vgpr_from_waves(program, program->num_waves);
      program->max_reg_demand.sgpr = get_addr_sgpr_from_waves(program, program->num_waves);
   }
}

void
live_var_analysis(Program* program)
{
   program->live.live_in.clear();
   program->live.memory.release();
   program->live.live_in.resize(program->blocks.size(), IDSet(program->live.memory));
   program->max_reg_demand = RegisterDemand();
   program->needs_vcc = program->gfx_level >= GFX10;

   live_ctx ctx;
   ctx.program = program;
   ctx.worklist = program->blocks.size() - 1;
   ctx.handled_once = program->blocks.size();

   /* this implementation assumes that the block idx corresponds to the block's position in
    * program->blocks vector */
   while (ctx.worklist >= 0) {
      process_live_temps_per_block(ctx, &program->blocks[ctx.worklist--]);
   }

   /* calculate the program's register demand and number of waves */
   if (program->progress < CompilationProgress::after_ra)
      update_vgpr_sgpr_demand(program, program->max_reg_demand);
}

} // namespace aco