#include "nir.h"
#include "nir_builder.h"
+#include "nir_deref.h"
/** @file nir_lower_io_arrays_to_elements.c
*
*/
static unsigned
-get_io_offset(nir_builder *b, nir_deref_var *deref, nir_variable *var,
- unsigned *element_index)
+get_io_offset(nir_builder *b, nir_deref_instr *deref, nir_variable *var,
+ unsigned *element_index, unsigned *xfb_offset,
+ nir_ssa_def **vertex_index)
{
- bool vs_in = (b->shader->info.stage == MESA_SHADER_VERTEX) &&
- (var->data.mode == nir_var_shader_in);
+ nir_deref_path path;
+ nir_deref_path_init(&path, deref, NULL);
- nir_deref *tail = &deref->deref;
+ assert(path.path[0]->deref_type == nir_deref_type_var);
+ nir_deref_instr **p = &path.path[1];
/* For per-vertex input arrays (i.e. geometry shader inputs), skip the
* outermost array index. Process the rest normally.
*/
if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
- tail = tail->child;
+ *vertex_index = nir_ssa_for_src(b, (*p)->arr.index, 1);
+ p++;
}
unsigned offset = 0;
- while (tail->child != NULL) {
- tail = tail->child;
+ *xfb_offset = 0;
+ for (; *p; p++) {
+ if ((*p)->deref_type == nir_deref_type_array) {
+ /* must not be indirect dereference */
+ unsigned index = nir_src_as_uint((*p)->arr.index);
- if (tail->deref_type == nir_deref_type_array) {
- nir_deref_array *deref_array = nir_deref_as_array(tail);
- assert(deref_array->deref_array_type != nir_deref_array_type_indirect);
+ unsigned size = glsl_count_attribute_slots((*p)->type, false);
+ offset += size * index;
- unsigned size = glsl_count_attribute_slots(tail->type, vs_in);
- offset += size * deref_array->base_offset;
+ xfb_offset += index * glsl_get_component_slots((*p)->type) * 4;
- unsigned num_elements = glsl_type_is_array(tail->type) ?
- glsl_get_aoa_size(tail->type) : 1;
+ unsigned num_elements = glsl_type_is_array((*p)->type) ?
+ glsl_get_aoa_size((*p)->type) : 1;
- num_elements *= glsl_type_is_matrix(glsl_without_array(tail->type)) ?
- glsl_get_matrix_columns(glsl_without_array(tail->type)) : 1;
+ num_elements *= glsl_type_is_matrix(glsl_without_array((*p)->type)) ?
+ glsl_get_matrix_columns(glsl_without_array((*p)->type)) : 1;
- *element_index += num_elements * deref_array->base_offset;
- } else if (tail->deref_type == nir_deref_type_struct) {
+ *element_index += num_elements * index;
+ } else if ((*p)->deref_type == nir_deref_type_struct) {
/* TODO: we could also add struct splitting support to this pass */
break;
}
}
+ nir_deref_path_finish(&path);
+
return offset;
}
return elements;
}
-static void
-create_array_deref(nir_intrinsic_instr *arr_intr,
- nir_intrinsic_instr *element_intr)
-{
- assert(arr_intr->variables[0]->deref.child);
-
- nir_deref *parent = &element_intr->variables[0]->deref;
- nir_deref_array *darr =
- nir_deref_as_array(arr_intr->variables[0]->deref.child);
- nir_deref_array *ndarr = nir_deref_array_create(parent);
-
- ndarr->deref.type = glsl_get_array_element(parent->type);
- ndarr->deref_array_type = darr->deref_array_type;
- ndarr->base_offset = darr->base_offset;
-
- if (ndarr->deref_array_type == nir_deref_array_type_indirect)
- nir_src_copy(&ndarr->indirect, &darr->indirect, parent);
-
- element_intr->variables[0]->deref.child = &ndarr->deref;
-}
-
static void
lower_array(nir_builder *b, nir_intrinsic_instr *intr, nir_variable *var,
struct hash_table *varyings)
nir_variable **elements =
get_array_elements(varyings, var, b->shader->info.stage);
+ nir_ssa_def *vertex_index = NULL;
unsigned elements_index = 0;
- unsigned io_offset = get_io_offset(b, intr->variables[0], var,
- &elements_index);
+ unsigned xfb_offset = 0;
+ unsigned io_offset = get_io_offset(b, nir_src_as_deref(intr->src[0]),
+ var, &elements_index, &xfb_offset,
+ &vertex_index);
nir_variable *element = elements[elements_index];
if (!element) {
element = nir_variable_clone(var, b->shader);
element->data.location = var->data.location + io_offset;
+ if (var->data.explicit_offset)
+ element->data.offset = var->data.offset + xfb_offset;
+
const struct glsl_type *type = glsl_without_array(element->type);
/* This pass also splits matrices so we need give them a new type. */
- if (glsl_type_is_matrix(type)) {
- type = glsl_vector_type(glsl_get_base_type(type),
- glsl_get_vector_elements(type));
- }
+ if (glsl_type_is_matrix(type))
+ type = glsl_get_column_type(type);
if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
- type = glsl_get_array_instance(type,
- glsl_get_length(element->type));
+ type = glsl_array_type(type, glsl_get_length(element->type),
+ glsl_get_explicit_stride(element->type));
}
element->type = type;
nir_shader_add_variable(b->shader, element);
}
+ nir_deref_instr *element_deref = nir_build_deref_var(b, element);
+
+ if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
+ assert(vertex_index);
+ element_deref = nir_build_deref_array(b, element_deref, vertex_index);
+ }
+
nir_intrinsic_instr *element_intr =
nir_intrinsic_instr_create(b->shader, intr->intrinsic);
element_intr->num_components = intr->num_components;
- element_intr->variables[0] = nir_deref_var_create(element_intr, element);
+ element_intr->src[0] = nir_src_for_ssa(&element_deref->dest.ssa);
- if (intr->intrinsic != nir_intrinsic_store_var) {
+ if (intr->intrinsic != nir_intrinsic_store_deref) {
nir_ssa_dest_init(&element_intr->instr, &element_intr->dest,
intr->num_components, intr->dest.ssa.bit_size, NULL);
- if (intr->intrinsic == nir_intrinsic_interp_var_at_offset ||
- intr->intrinsic == nir_intrinsic_interp_var_at_sample) {
- nir_src_copy(&element_intr->src[0], &intr->src[0],
+ if (intr->intrinsic == nir_intrinsic_interp_deref_at_offset ||
+ intr->intrinsic == nir_intrinsic_interp_deref_at_sample ||
+ intr->intrinsic == nir_intrinsic_interp_deref_at_vertex) {
+ nir_src_copy(&element_intr->src[1], &intr->src[1],
&element_intr->instr);
}
} else {
nir_intrinsic_set_write_mask(element_intr,
nir_intrinsic_write_mask(intr));
- nir_src_copy(&element_intr->src[0], &intr->src[0],
+ nir_src_copy(&element_intr->src[1], &intr->src[1],
&element_intr->instr);
}
- if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
- create_array_deref(intr, element_intr);
- }
-
nir_builder_instr_insert(b, &element_intr->instr);
/* Remove the old load intrinsic */
}
static bool
-deref_has_indirect(nir_builder *b, nir_variable *var, nir_deref_var *deref)
+deref_has_indirect(nir_builder *b, nir_variable *var, nir_deref_path *path)
{
- nir_deref *tail = &deref->deref;
+ assert(path->path[0]->deref_type == nir_deref_type_var);
+ nir_deref_instr **p = &path->path[1];
if (nir_is_per_vertex_io(var, b->shader->info.stage)) {
- tail = tail->child;
+ p++;
}
- for (tail = tail->child; tail; tail = tail->child) {
- if (tail->deref_type != nir_deref_type_array)
+ for (; *p; p++) {
+ if ((*p)->deref_type != nir_deref_type_array)
continue;
- nir_deref_array *arr = nir_deref_as_array(tail);
- if (arr->deref_array_type == nir_deref_array_type_indirect)
+ if (!nir_src_is_const((*p)->arr.index))
return true;
}
* indirect indexing.
*/
static void
-create_indirects_mask(nir_shader *shader, uint64_t *indirects,
- uint64_t *patch_indirects, nir_variable_mode mode)
+create_indirects_mask(nir_shader *shader,
+ BITSET_WORD *indirects, nir_variable_mode mode)
{
nir_foreach_function(function, shader) {
if (function->impl) {
nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
- if (intr->intrinsic != nir_intrinsic_load_var &&
- intr->intrinsic != nir_intrinsic_store_var &&
- intr->intrinsic != nir_intrinsic_interp_var_at_centroid &&
- intr->intrinsic != nir_intrinsic_interp_var_at_sample &&
- intr->intrinsic != nir_intrinsic_interp_var_at_offset)
+ if (intr->intrinsic != nir_intrinsic_load_deref &&
+ intr->intrinsic != nir_intrinsic_store_deref &&
+ intr->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
+ intr->intrinsic != nir_intrinsic_interp_deref_at_sample &&
+ intr->intrinsic != nir_intrinsic_interp_deref_at_offset &&
+ intr->intrinsic != nir_intrinsic_interp_deref_at_vertex)
continue;
- nir_variable *var = intr->variables[0]->var;
-
- if (var->data.mode != mode)
+ nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
+ if (deref->mode != mode)
continue;
- uint64_t loc_mask = ((uint64_t)1) << var->data.location;
- if (var->data.patch) {
- if (deref_has_indirect(&b, var, intr->variables[0]))
- patch_indirects[var->data.location_frac] |= loc_mask;
- } else {
- if (deref_has_indirect(&b, var, intr->variables[0]))
- indirects[var->data.location_frac] |= loc_mask;
- }
+ nir_variable *var = nir_deref_instr_get_variable(deref);
+
+ nir_deref_path path;
+ nir_deref_path_init(&path, deref, NULL);
+
+ int loc = var->data.location * 4 + var->data.location_frac;
+ if (deref_has_indirect(&b, var, &path))
+ BITSET_SET(indirects, loc);
+
+ nir_deref_path_finish(&path);
}
}
}
static void
lower_io_arrays_to_elements(nir_shader *shader, nir_variable_mode mask,
- uint64_t *indirects, uint64_t *patch_indirects,
+ BITSET_WORD *indirects,
struct hash_table *varyings,
bool after_cross_stage_opts)
{
nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
- if (intr->intrinsic != nir_intrinsic_load_var &&
- intr->intrinsic != nir_intrinsic_store_var &&
- intr->intrinsic != nir_intrinsic_interp_var_at_centroid &&
- intr->intrinsic != nir_intrinsic_interp_var_at_sample &&
- intr->intrinsic != nir_intrinsic_interp_var_at_offset)
+ if (intr->intrinsic != nir_intrinsic_load_deref &&
+ intr->intrinsic != nir_intrinsic_store_deref &&
+ intr->intrinsic != nir_intrinsic_interp_deref_at_centroid &&
+ intr->intrinsic != nir_intrinsic_interp_deref_at_sample &&
+ intr->intrinsic != nir_intrinsic_interp_deref_at_offset &&
+ intr->intrinsic != nir_intrinsic_interp_deref_at_vertex)
+ continue;
+
+ nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
+ if (!(deref->mode & mask))
continue;
- nir_variable *var = intr->variables[0]->var;
+ nir_variable *var = nir_deref_instr_get_variable(deref);
+
+ /* Drivers assume compact arrays are, in fact, arrays. */
+ if (var->data.compact)
+ continue;
+
+ /* Per-view variables are expected to remain arrays. */
+ if (var->data.per_view)
+ continue;
/* Skip indirects */
- uint64_t loc_mask = ((uint64_t)1) << var->data.location;
- if (var->data.patch) {
- if (patch_indirects[var->data.location_frac] & loc_mask)
- continue;
- } else {
- if (indirects[var->data.location_frac] & loc_mask)
- continue;
- }
+ int loc = var->data.location * 4 + var->data.location_frac;
+ if (BITSET_TEST(indirects, loc))
+ continue;
nir_variable_mode mode = var->data.mode;
* TODO: Add support for struct splitting.
*/
if ((!glsl_type_is_array(type) && !glsl_type_is_matrix(type))||
- glsl_type_is_struct(glsl_without_array(type)))
+ glsl_type_is_struct_or_ifc(glsl_without_array(type)))
continue;
/* Skip builtins */
continue;
switch (intr->intrinsic) {
- case nir_intrinsic_interp_var_at_centroid:
- case nir_intrinsic_interp_var_at_sample:
- case nir_intrinsic_interp_var_at_offset:
- case nir_intrinsic_load_var:
- case nir_intrinsic_store_var:
+ case nir_intrinsic_interp_deref_at_centroid:
+ case nir_intrinsic_interp_deref_at_sample:
+ case nir_intrinsic_interp_deref_at_offset:
+ case nir_intrinsic_interp_deref_at_vertex:
+ case nir_intrinsic_load_deref:
+ case nir_intrinsic_store_deref:
if ((mask & nir_var_shader_in && mode == nir_var_shader_in) ||
(mask & nir_var_shader_out && mode == nir_var_shader_out))
lower_array(&b, intr, var, varyings);
}
void
-nir_lower_io_arrays_to_elements_no_indirects(nir_shader *shader)
+nir_lower_io_arrays_to_elements_no_indirects(nir_shader *shader,
+ bool outputs_only)
{
- struct hash_table *split_inputs =
- _mesa_hash_table_create(NULL, _mesa_hash_pointer,
- _mesa_key_pointer_equal);
- struct hash_table *split_outputs =
- _mesa_hash_table_create(NULL, _mesa_hash_pointer,
- _mesa_key_pointer_equal);
+ struct hash_table *split_inputs = _mesa_pointer_hash_table_create(NULL);
+ struct hash_table *split_outputs = _mesa_pointer_hash_table_create(NULL);
- uint64_t indirects[4] = {0}, patch_indirects[4] = {0};
+ BITSET_DECLARE(indirects, 4 * VARYING_SLOT_TESS_MAX) = {0};
- lower_io_arrays_to_elements(shader, nir_var_shader_out, indirects,
- patch_indirects, split_outputs, true);
+ lower_io_arrays_to_elements(shader, nir_var_shader_out,
+ indirects, split_outputs, true);
- lower_io_arrays_to_elements(shader, nir_var_shader_in, indirects,
- patch_indirects, split_inputs, true);
+ if (!outputs_only) {
+ lower_io_arrays_to_elements(shader, nir_var_shader_in,
+ indirects, split_inputs, true);
- /* Remove old input from the shaders inputs list */
- struct hash_entry *entry;
- hash_table_foreach(split_inputs, entry) {
- nir_variable *var = (nir_variable *) entry->key;
- exec_node_remove(&var->node);
+ /* Remove old input from the shaders inputs list */
+ hash_table_foreach(split_inputs, entry) {
+ nir_variable *var = (nir_variable *) entry->key;
+ exec_node_remove(&var->node);
- free(entry->data);
+ free(entry->data);
+ }
}
/* Remove old output from the shaders outputs list */
_mesa_hash_table_destroy(split_inputs, NULL);
_mesa_hash_table_destroy(split_outputs, NULL);
+
+ nir_remove_dead_derefs(shader);
}
void
nir_lower_io_arrays_to_elements(nir_shader *producer, nir_shader *consumer)
{
- struct hash_table *split_inputs =
- _mesa_hash_table_create(NULL, _mesa_hash_pointer,
- _mesa_key_pointer_equal);
- struct hash_table *split_outputs =
- _mesa_hash_table_create(NULL, _mesa_hash_pointer,
- _mesa_key_pointer_equal);
+ struct hash_table *split_inputs = _mesa_pointer_hash_table_create(NULL);
+ struct hash_table *split_outputs = _mesa_pointer_hash_table_create(NULL);
- uint64_t indirects[4] = {0}, patch_indirects[4] = {0};
- create_indirects_mask(producer, indirects, patch_indirects,
- nir_var_shader_out);
- create_indirects_mask(consumer, indirects, patch_indirects,
- nir_var_shader_in);
+ BITSET_DECLARE(indirects, 4 * VARYING_SLOT_TESS_MAX) = {0};
- lower_io_arrays_to_elements(producer, nir_var_shader_out, indirects,
- patch_indirects, split_outputs, false);
+ create_indirects_mask(producer, indirects, nir_var_shader_out);
+ create_indirects_mask(consumer, indirects, nir_var_shader_in);
- lower_io_arrays_to_elements(consumer, nir_var_shader_in, indirects,
- patch_indirects, split_inputs, false);
+ lower_io_arrays_to_elements(producer, nir_var_shader_out,
+ indirects, split_outputs, false);
+
+ lower_io_arrays_to_elements(consumer, nir_var_shader_in,
+ indirects, split_inputs, false);
/* Remove old input from the shaders inputs list */
- struct hash_entry *entry;
hash_table_foreach(split_inputs, entry) {
nir_variable *var = (nir_variable *) entry->key;
exec_node_remove(&var->node);
_mesa_hash_table_destroy(split_inputs, NULL);
_mesa_hash_table_destroy(split_outputs, NULL);
+
+ nir_remove_dead_derefs(producer);
+ nir_remove_dead_derefs(consumer);
}
projects / mesa.git / blobdiff