aco: only reserve sgprs for vcc if it's used

pipeline-db (Vega):

Totals:
SGPRS: 5186302 -> 5075616 (-2.13 %)
VGPRS: 3704580 -> 3704580 (0.00 %)
Spilled SGPRs: 144859 -> 144859 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Scratch size: 4124 -> 4124 (0.00 %) dwords per thread
Code Size: 247315944 -> 247315944 (0.00 %) bytes
LDS: 1311 -> 1311 (0.00 %) blocks
Max Waves: 674560 -> 674562 (0.00 %)

Totals from affected shaders:
SGPRS: 536992 -> 426306 (-20.61 %)
VGPRS: 356404 -> 356404 (0.00 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 8498748 -> 8498748 (0.00 %) bytes
LDS: 8 -> 8 (0.00 %) blocks
Max Waves: 113832 -> 113834 (0.00 %)

There are some small code size changes in a few RotTR shaders and a small
increase in max_waves in two Detroit: Become Human shaders.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3906>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3906>

src/amd/compiler/aco_instruction_selection_setup.cpp

@@ -1007,9 +1007,6 @@ setup_isel_context(Program* program,
       program->sgpr_limit = 104;
    }
 
-   /* TODO: we don't have to allocate VCC if we don't need it */
-   program->needs_vcc = true;
-
    calc_min_waves(program);
    program->vgpr_limit = get_addr_vgpr_from_waves(program, program->min_waves);
    program->sgpr_limit = get_addr_sgpr_from_waves(program, program->min_waves);

src/amd/compiler/aco_live_var_analysis.cpp

@@ -87,6 +87,8 @@ void process_live_temps_per_block(Program *program, live& lives, Block* block,
          if (!definition.isTemp()) {
             continue;
          }
+         if ((definition.isFixed() || definition.hasHint()) && definition.physReg() == vcc)
+            program->needs_vcc = true;
 
          const Temp temp = definition.getTemp();
          size_t n = 0;
@@ -120,9 +122,10 @@ void process_live_temps_per_block(Program *program, live& lives, Block* block,
          for (unsigned i = 0; i < insn->operands.size(); ++i)
          {
             Operand& operand = insn->operands[i];
-            if (!operand.isTemp()) {
+            if (!operand.isTemp())
                continue;
-            }
+            if (operand.isFixed() && operand.physReg() == vcc)
+               program->needs_vcc = true;
             const Temp temp = operand.getTemp();
             const bool inserted = temp.is_linear()
                                 ? live_sgprs.insert(temp).second
@@ -161,6 +164,8 @@ void process_live_temps_per_block(Program *program, live& lives, Block* block,
       assert(is_phi(insn));
       assert(insn->definitions.size() == 1 && insn->definitions[0].isTemp());
       Definition& definition = insn->definitions[0];
+      if ((definition.isFixed() || definition.hasHint()) && definition.physReg() == vcc)
+         program->needs_vcc = true;
       const Temp temp = definition.getTemp();
       size_t n = 0;
 
@@ -205,9 +210,10 @@ void process_live_temps_per_block(Program *program, live& lives, Block* block,
                                    : block->linear_preds;
       for (unsigned i = 0; i < preds.size(); ++i) {
          Operand &operand = insn->operands[i];
-         if (!operand.isTemp()) {
+         if (!operand.isTemp())
             continue;
-         }
+         if (operand.isFixed() && operand.physReg() == vcc)
+            program->needs_vcc = true;
          /* check if we changed an already processed block */
          const bool inserted = live_temps[preds[i]].insert(operand.getTemp()).second;
          if (inserted) {
@@ -364,6 +370,8 @@ live live_var_analysis(Program* program,
    std::vector<uint16_t> phi_sgpr_ops(program->blocks.size());
    RegisterDemand new_demand;
 
+   program->needs_vcc = false;
+
    /* this implementation assumes that the block idx corresponds to the block's position in program->blocks vector */
    for (Block& block : program->blocks)
       worklist.insert(block.index);

src/amd/compiler/aco_validate.cpp

@@ -390,6 +390,8 @@ bool validate_ra(Program *program, const struct radv_nir_compiler_options *optio
             if ((op.getTemp().type() == RegType::vgpr && op.physReg() + op.size() > 256 + program->config->num_vgprs) ||
                 (op.getTemp().type() == RegType::sgpr && op.physReg() + op.size() > program->config->num_sgprs && op.physReg() < program->sgpr_limit))
                err |= ra_fail(output, loc, assignments.at(op.tempId()).firstloc, "Operand %d has an out-of-bounds register assignment", i);
+            if (op.physReg() == vcc && !program->needs_vcc)
+               err |= ra_fail(output, loc, Location(), "Operand %d fixed to vcc but needs_vcc=false", i);
             if (!assignments[op.tempId()].firstloc.block)
                assignments[op.tempId()].firstloc = loc;
             if (!assignments[op.tempId()].defloc.block)
@@ -407,6 +409,8 @@ bool validate_ra(Program *program, const struct radv_nir_compiler_options *optio
             if ((def.getTemp().type() == RegType::vgpr && def.physReg() + def.size() > 256 + program->config->num_vgprs) ||
                 (def.getTemp().type() == RegType::sgpr && def.physReg() + def.size() > program->config->num_sgprs && def.physReg() < program->sgpr_limit))
                err |= ra_fail(output, loc, assignments.at(def.tempId()).firstloc, "Definition %d has an out-of-bounds register assignment", i);
+            if (def.physReg() == vcc && !program->needs_vcc)
+               err |= ra_fail(output, loc, Location(), "Definition %d fixed to vcc but needs_vcc=false", i);
             if (!assignments[def.tempId()].firstloc.block)
                assignments[def.tempId()].firstloc = loc;
             assignments[def.tempId()].defloc = loc;