glsl: add KHR_shader_subgroup_clustered builtin functions

Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30610>

src/compiler/glsl/builtin_functions.cpp

@@ -1093,6 +1093,18 @@ subgroup_arithmetic_and_fp64(const _mesa_glsl_parse_state *state)
    return subgroup_arithmetic(state) && fp64(state);
 }
 
+static bool
+subgroup_clustered(const _mesa_glsl_parse_state *state)
+{
+   return state->KHR_shader_subgroup_clustered_enable;
+}
+
+static bool
+subgroup_clustered_and_fp64(const _mesa_glsl_parse_state *state)
+{
+   return subgroup_clustered(state) && fp64(state);
+}
+
 /** @} */
 
 /******************************************************************************/
@@ -1526,6 +1538,11 @@ private:
    ir_function_signature *_subgroup_arithmetic(const glsl_type *type,
                                                const char *intrinsic_name);
 
+   ir_function_signature *_subgroup_clustered_intrinsic(const glsl_type *type,
+                                                        enum ir_intrinsic_id id);
+   ir_function_signature *_subgroup_clustered(const glsl_type *type,
+                                              const char *intrinsic_name);
+
 #undef B0
 #undef B1
 #undef B2
@@ -1989,6 +2006,8 @@ builtin_builder::create_intrinsics()
    SUBGROUP_ARITH_INTRINSICS(arithmetic, reduce);
    SUBGROUP_ARITH_INTRINSICS(arithmetic, inclusive);
    SUBGROUP_ARITH_INTRINSICS(arithmetic, exclusive);
+
+   SUBGROUP_ARITH_INTRINSICS(clustered, clustered);
 }
 
 /**
@@ -5926,6 +5945,8 @@ builtin_builder::create_builtins()
    SUBGROUP_ARITH(arithmetic, Inclusive, inclusive);
    SUBGROUP_ARITH(arithmetic, Exclusive, exclusive);
 
+   SUBGROUP_ARITH(clustered, Clustered, clustered);
+
 #undef F
 #undef FI
 #undef FIUDHF_VEC
@@ -9350,6 +9371,35 @@ builtin_builder::_subgroup_arithmetic(const glsl_type *type, const char *intrins
    return sig;
 }
 
+ir_function_signature *
+builtin_builder::_subgroup_clustered_intrinsic(const glsl_type *type, enum ir_intrinsic_id id)
+{
+   ir_variable *value = in_var(type, "value");
+   ir_variable *size =
+      new(mem_ctx) ir_variable(&glsl_type_builtin_uint, "clusterSize", ir_var_const_in);
+
+   MAKE_INTRINSIC(type, id,
+                  glsl_type_is_double(type) ? subgroup_clustered_and_fp64 : subgroup_clustered,
+                  2, value, size);
+   return sig;
+}
+
+ir_function_signature *
+builtin_builder::_subgroup_clustered(const glsl_type *type, const char *intrinsic_name)
+{
+   ir_variable *value = in_var(type, "value");
+   ir_variable *size =
+      new(mem_ctx) ir_variable(&glsl_type_builtin_uint, "clusterSize", ir_var_const_in);
+
+   MAKE_SIG(type, glsl_type_is_double(type) ? subgroup_clustered_and_fp64 : subgroup_clustered,
+            2, value, size);
+
+   ir_variable *retval = body.make_temp(type, "retval");
+   body.emit(call(shader->symbols->get_function(intrinsic_name), retval, sig->parameters));
+   body.emit(ret(retval));
+   return sig;
+}
+
 /** @} */
 
 /******************************************************************************/

src/compiler/glsl/glsl_to_nir.cpp

@@ -856,6 +856,7 @@ get_reduction_op(enum ir_intrinsic_id id, const glsl_type *type)
    case ir_intrinsic_reduce_##op: \
    case ir_intrinsic_inclusive_##op: \
    case ir_intrinsic_exclusive_##op: \
+   case ir_intrinsic_clustered_##op: \
       return CONV_OP(op);
 
    switch (id) {
@@ -1147,6 +1148,13 @@ nir_visitor::visit(ir_call *ir)
       case ir_intrinsic_reduce_and:
       case ir_intrinsic_reduce_or:
       case ir_intrinsic_reduce_xor:
+      case ir_intrinsic_clustered_add:
+      case ir_intrinsic_clustered_mul:
+      case ir_intrinsic_clustered_min:
+      case ir_intrinsic_clustered_max:
+      case ir_intrinsic_clustered_and:
+      case ir_intrinsic_clustered_or:
+      case ir_intrinsic_clustered_xor:
          op = nir_intrinsic_reduce;
          break;
       case ir_intrinsic_inclusive_add:
@@ -1548,9 +1556,18 @@ nir_visitor::visit(ir_call *ir)
                       glsl_get_bit_size(type));
          instr->num_components = instr->def.num_components;
 
-         ir_rvalue *value = (ir_rvalue *)ir->actual_parameters.get_head();
+         exec_node *param = ir->actual_parameters.get_head();
+         ir_rvalue *value = ((ir_instruction *)param)->as_rvalue();
          instr->src[0] = nir_src_for_ssa(evaluate_rvalue(value));
 
+         param = param->get_next();
+         if (!param->is_tail_sentinel()) {
+            ir_constant *size = ((ir_instruction *)param)->as_constant();
+            assert(size);
+
+            nir_intrinsic_set_cluster_size(instr, size->get_uint_component(0));
+         }
+
          nir_intrinsic_set_reduction_op(instr, get_reduction_op(ir->callee->intrinsic_id, type));
 
          nir_builder_instr_insert(&b, &instr->instr);

src/compiler/glsl/ir.h

@@ -1175,6 +1175,14 @@ enum ir_intrinsic_id {
    ir_intrinsic_exclusive_and,
    ir_intrinsic_exclusive_or,
    ir_intrinsic_exclusive_xor,
+
+   ir_intrinsic_clustered_add,
+   ir_intrinsic_clustered_mul,
+   ir_intrinsic_clustered_min,
+   ir_intrinsic_clustered_max,
+   ir_intrinsic_clustered_and,
+   ir_intrinsic_clustered_or,
+   ir_intrinsic_clustered_xor,
 };
 
 /*@{*/