intel/brw: Add SHADER_OPCODE_BALLOT

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31052>

src/intel/compiler/brw_eu_defines.h

@@ -437,6 +437,13 @@ enum opcode {
     */
    SHADER_OPCODE_VOTE_EQUAL,
 
+   /* Produces a mask from the boolean value from all channels, and broadcast
+    * the result to all channels.
+    *
+    * Source 0: Boolean value.
+    */
+   SHADER_OPCODE_BALLOT,
+
    /* Select between src0 and src1 based on channel enables.
     *
     * This instruction copies src0 into the enabled channels of the

src/intel/compiler/brw_fs.cpp

@@ -324,6 +324,7 @@ fs_inst::can_do_source_mods(const struct intel_device_info *devinfo) const
    case SHADER_OPCODE_VOTE_ANY:
    case SHADER_OPCODE_VOTE_ALL:
    case SHADER_OPCODE_VOTE_EQUAL:
+   case SHADER_OPCODE_BALLOT:
       return false;
    default:
       return true;

src/intel/compiler/brw_fs_copy_propagation.cpp

@@ -1216,6 +1216,7 @@ try_constant_propagate_value(brw_reg val, brw_reg_type dst_type,
    case BRW_OPCODE_LRP:
    case FS_OPCODE_PACK_HALF_2x16_SPLIT:
    case SHADER_OPCODE_SHUFFLE:
+   case SHADER_OPCODE_BALLOT:
       inst->src[arg] = val;
       progress = true;
       break;

src/intel/compiler/brw_fs_nir.cpp

@@ -6627,25 +6627,8 @@ fs_nir_emit_intrinsic(nir_to_brw_state &ntb,
          dest.type = BRW_TYPE_UD;
       }
 
-      /* Implement a fast-path for ballot(true). */
+      const brw_reg value = get_nir_src(ntb, instr->src[0]);
-      if (nir_src_is_const(instr->src[0]) &&
+      bld.emit(SHADER_OPCODE_BALLOT, dest, value);
-          nir_src_as_bool(instr->src[0])) {
-         brw_reg tmp = bld.vgrf(BRW_TYPE_UD);
-         bld.exec_all().emit(SHADER_OPCODE_LOAD_LIVE_CHANNELS, tmp);
-         bld.MOV(dest, brw_reg(component(tmp, 0)));
-         break;
-      }
-
-      const brw_reg value = retype(get_nir_src(ntb, instr->src[0]),
-                                  BRW_TYPE_UD);
-      struct brw_reg flag = brw_flag_reg(0, 0);
-
-      if (s.dispatch_width == 32)
-         flag.type = BRW_TYPE_UD;
-
-      bld.exec_all().group(1, 0).MOV(flag, retype(brw_imm_ud(0u), flag.type));
-      bld.CMP(bld.null_reg_ud(), value, brw_imm_ud(0u), BRW_CONDITIONAL_NZ);
-      bld.MOV(dest, flag);
       break;
    }
 

src/intel/compiler/brw_fs_validate.cpp

@@ -237,6 +237,7 @@ brw_validate_instruction_phase(const fs_visitor &s, fs_inst *inst)
    case SHADER_OPCODE_VOTE_ANY:
    case SHADER_OPCODE_VOTE_ALL:
    case SHADER_OPCODE_VOTE_EQUAL:
+   case SHADER_OPCODE_BALLOT:
       invalid_from = BRW_SHADER_PHASE_AFTER_EARLY_LOWERING;
       break;
 

src/intel/compiler/brw_lower_subgroup_ops.cpp

@@ -511,6 +511,34 @@ brw_lower_vote(fs_visitor &s, bblock_t *block, fs_inst *inst)
    return true;
 }
 
+static bool
+brw_lower_ballot(fs_visitor &s, bblock_t *block, fs_inst *inst)
+{
+   const fs_builder bld(&s, block, inst);
+
+   brw_reg value = retype(inst->src[0], BRW_TYPE_UD);
+   brw_reg dst = inst->dst;
+
+   if (value.file == IMM) {
+      /* Implement a fast-path for ballot(true). */
+      if (!value.is_zero()) {
+         brw_reg tmp = bld.vgrf(BRW_TYPE_UD);
+         bld.exec_all().emit(SHADER_OPCODE_LOAD_LIVE_CHANNELS, tmp);
+         bld.MOV(dst, brw_reg(component(tmp, 0)));
+      } else {
+         brw_reg zero = retype(brw_imm_uq(0), dst.type);
+         bld.MOV(dst, zero);
+      }
+   } else {
+      brw_reg flag = brw_fill_flag(bld, 0);
+      bld.CMP(bld.null_reg_ud(), value, brw_imm_ud(0u), BRW_CONDITIONAL_NZ);
+      bld.MOV(dst, flag);
+   }
+
+   inst->remove(block);
+   return true;
+}
+
 bool
 brw_fs_lower_subgroup_ops(fs_visitor &s)
 {
@@ -533,6 +561,10 @@ brw_fs_lower_subgroup_ops(fs_visitor &s)
          progress |= brw_lower_vote(s, block, inst);
          break;
 
+      case SHADER_OPCODE_BALLOT:
+         progress |= brw_lower_ballot(s, block, inst);
+         break;
+
       default:
          /* Nothing to do. */
          break;

src/intel/compiler/brw_print.cpp

@@ -303,6 +303,8 @@ brw_instruction_name(const struct brw_isa_info *isa, enum opcode op)
       return "vote_all";
    case SHADER_OPCODE_VOTE_EQUAL:
       return "vote_equal";
+   case SHADER_OPCODE_BALLOT:
+      return "ballot";
    }
 
    unreachable("not reached");