/*
* This lowering pass converts references to input/output variables with
* loads/stores to actual input/output intrinsics.
- *
- * NOTE: This pass really only works for scalar backends at the moment due
- * to the way it packes the input/output data.
*/
#include "nir.h"
struct lower_io_state {
void *mem_ctx;
+ bool is_scalar;
};
+static int
+type_size_vec4(const struct glsl_type *type)
+{
+ unsigned int i;
+ int size;
+
+ switch (glsl_get_base_type(type)) {
+ case GLSL_TYPE_UINT:
+ case GLSL_TYPE_INT:
+ case GLSL_TYPE_FLOAT:
+ case GLSL_TYPE_BOOL:
+ if (glsl_type_is_matrix(type)) {
+ return glsl_get_matrix_columns(type);
+ } else {
+ return 1;
+ }
+ case GLSL_TYPE_ARRAY:
+ return type_size_vec4(glsl_get_array_element(type)) * glsl_get_length(type);
+ case GLSL_TYPE_STRUCT:
+ size = 0;
+ for (i = 0; i < glsl_get_length(type); i++) {
+ size += type_size_vec4(glsl_get_struct_field(type, i));
+ }
+ return size;
+ case GLSL_TYPE_SUBROUTINE:
+ return 1;
+ case GLSL_TYPE_SAMPLER:
+ return 0;
+ case GLSL_TYPE_ATOMIC_UINT:
+ return 0;
+ case GLSL_TYPE_IMAGE:
+ case GLSL_TYPE_VOID:
+ case GLSL_TYPE_DOUBLE:
+ case GLSL_TYPE_ERROR:
+ case GLSL_TYPE_INTERFACE:
+ unreachable("not reached");
+ }
+
+ return 0;
+}
+
static unsigned
-type_size(const struct glsl_type *type)
+type_size_scalar(const struct glsl_type *type)
{
unsigned int size, i;
case GLSL_TYPE_BOOL:
return glsl_get_components(type);
case GLSL_TYPE_ARRAY:
- return type_size(glsl_get_array_element(type)) * glsl_get_length(type);
+ return type_size_scalar(glsl_get_array_element(type)) * glsl_get_length(type);
case GLSL_TYPE_STRUCT:
size = 0;
for (i = 0; i < glsl_get_length(type); i++) {
- size += type_size(glsl_get_struct_field(type, i));
+ size += type_size_scalar(glsl_get_struct_field(type, i));
}
return size;
+ case GLSL_TYPE_SUBROUTINE:
+ return 1;
case GLSL_TYPE_SAMPLER:
return 0;
case GLSL_TYPE_ATOMIC_UINT:
return 0;
case GLSL_TYPE_VOID:
case GLSL_TYPE_ERROR:
+ case GLSL_TYPE_DOUBLE:
unreachable("not reached");
}
return 0;
}
-static void
-assign_var_locations(struct hash_table *ht, unsigned *size)
+static unsigned
+type_size(const struct glsl_type *type, bool is_scalar)
{
- unsigned location = 0;
+ if (is_scalar)
+ return type_size_scalar(type);
+ else
+ return type_size_vec4(type);
+}
- struct hash_entry *entry;
- hash_table_foreach(ht, entry) {
- nir_variable *var = (nir_variable *) entry->data;
+void
+nir_assign_var_locations(struct exec_list *var_list, unsigned *size, bool is_scalar)
+{
+ unsigned location = 0;
+ foreach_list_typed(nir_variable, var, node, var_list) {
/*
* UBO's have their own address spaces, so don't count them towards the
* number of global uniforms
*/
- if (var->data.mode == nir_var_uniform && var->interface_type != NULL)
+ if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
+ var->interface_type != NULL)
continue;
var->data.driver_location = location;
- location += type_size(var->type);
+ location += type_size(var->type, is_scalar);
}
*size = location;
}
-static void
-assign_var_locations_shader(nir_shader *shader)
-{
- assign_var_locations(shader->inputs, &shader->num_inputs);
- assign_var_locations(shader->outputs, &shader->num_outputs);
- assign_var_locations(shader->uniforms, &shader->num_uniforms);
-}
-
static bool
deref_has_indirect(nir_deref_var *deref)
{
return false;
}
+static bool
+mark_indirect_uses_block(nir_block *block, void *void_state)
+{
+ struct set *indirect_set = void_state;
+
+ nir_foreach_instr(block, instr) {
+ if (instr->type != nir_instr_type_intrinsic)
+ continue;
+
+ nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
+
+ for (unsigned i = 0;
+ i < nir_intrinsic_infos[intrin->intrinsic].num_variables; i++) {
+ if (deref_has_indirect(intrin->variables[i]))
+ _mesa_set_add(indirect_set, intrin->variables[i]->var);
+ }
+ }
+
+ return true;
+}
+
+/* Identical to nir_assign_var_locations_packed except that it assigns
+ * locations to the variables that are used 100% directly first and then
+ * assigns locations to variables that are used indirectly.
+ */
+void
+nir_assign_var_locations_direct_first(nir_shader *shader,
+ struct exec_list *var_list,
+ unsigned *direct_size,
+ unsigned *size,
+ bool is_scalar)
+{
+ struct set *indirect_set = _mesa_set_create(NULL, _mesa_hash_pointer,
+ _mesa_key_pointer_equal);
+
+ nir_foreach_overload(shader, overload) {
+ if (overload->impl)
+ nir_foreach_block(overload->impl, mark_indirect_uses_block,
+ indirect_set);
+ }
+
+ unsigned location = 0;
+
+ foreach_list_typed(nir_variable, var, node, var_list) {
+ if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
+ var->interface_type != NULL)
+ continue;
+
+ if (_mesa_set_search(indirect_set, var))
+ continue;
+
+ var->data.driver_location = location;
+ location += type_size(var->type, is_scalar);
+ }
+
+ *direct_size = location;
+
+ foreach_list_typed(nir_variable, var, node, var_list) {
+ if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
+ var->interface_type != NULL)
+ continue;
+
+ if (!_mesa_set_search(indirect_set, var))
+ continue;
+
+ var->data.driver_location = location;
+ location += type_size(var->type, is_scalar);
+ }
+
+ *size = location;
+
+ _mesa_set_destroy(indirect_set, NULL);
+}
+
static unsigned
get_io_offset(nir_deref_var *deref, nir_instr *instr, nir_src *indirect,
struct lower_io_state *state)
if (tail->deref_type == nir_deref_type_array) {
nir_deref_array *deref_array = nir_deref_as_array(tail);
- unsigned size = type_size(tail->type);
+ unsigned size = type_size(tail->type, state->is_scalar);
base_offset += size * deref_array->base_offset;
nir_op_imul);
mul->src[0].src.is_ssa = true;
mul->src[0].src.ssa = &load_const->def;
- mul->src[1].src = nir_src_copy(deref_array->indirect,
- state->mem_ctx);
+ nir_src_copy(&mul->src[1].src, &deref_array->indirect,
+ state->mem_ctx);
mul->dest.write_mask = 1;
- mul->dest.dest.is_ssa = true;
- nir_ssa_def_init(&mul->instr, &mul->dest.dest.ssa, 1, NULL);
+ nir_ssa_dest_init(&mul->instr, &mul->dest.dest, 1, NULL);
nir_instr_insert_before(instr, &mul->instr);
if (found_indirect) {
add->src[1].src.is_ssa = true;
add->src[1].src.ssa = &mul->dest.dest.ssa;
add->dest.write_mask = 1;
- add->dest.dest.is_ssa = true;
- nir_ssa_def_init(&add->instr, &add->dest.dest.ssa, 1, NULL);
+ nir_ssa_dest_init(&add->instr, &add->dest.dest, 1, NULL);
nir_instr_insert_before(instr, &add->instr);
indirect->is_ssa = true;
nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
for (unsigned i = 0; i < deref_struct->index; i++)
- base_offset += type_size(glsl_get_struct_field(parent_type, i));
+ base_offset += type_size(glsl_get_struct_field(parent_type, i),
+ state->is_scalar);
}
}
offset += intrin->variables[0]->var->data.driver_location;
load->const_index[0] = offset;
- load->const_index[1] = 1;
if (has_indirect)
load->src[0] = indirect;
if (intrin->dest.is_ssa) {
- load->dest.is_ssa = true;
- nir_ssa_def_init(&load->instr, &load->dest.ssa,
- intrin->num_components, NULL);
+ nir_ssa_dest_init(&load->instr, &load->dest,
+ intrin->num_components, NULL);
nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
nir_src_for_ssa(&load->dest.ssa),
state->mem_ctx);
} else {
- load->dest = nir_dest_copy(intrin->dest, state->mem_ctx);
+ nir_dest_copy(&load->dest, &intrin->dest, state->mem_ctx);
}
nir_instr_insert_before(&intrin->instr, &load->instr);
offset += intrin->variables[0]->var->data.driver_location;
store->const_index[0] = offset;
- store->const_index[1] = 1;
- store->src[0] = nir_src_copy(intrin->src[0], state->mem_ctx);
+ nir_src_copy(&store->src[0], &intrin->src[0], state->mem_ctx);
if (has_indirect)
store->src[1] = indirect;
}
static void
-nir_lower_io_impl(nir_function_impl *impl)
+nir_lower_io_impl(nir_function_impl *impl, bool is_scalar)
{
struct lower_io_state state;
state.mem_ctx = ralloc_parent(impl);
+ state.is_scalar = is_scalar;
nir_foreach_block(impl, nir_lower_io_block, &state);
}
void
-nir_lower_io(nir_shader *shader)
+nir_lower_io(nir_shader *shader, bool is_scalar)
{
- assign_var_locations_shader(shader);
-
nir_foreach_overload(shader, overload) {
if (overload->impl)
- nir_lower_io_impl(overload->impl);
+ nir_lower_io_impl(overload->impl, is_scalar);
}
}