static nir_register *
get_reg_for_deref(nir_deref_var *deref, struct locals_to_regs_state *state)
{
uint32_t hash = hash_deref(deref);
struct hash_entry *entry =
_mesa_hash_table_search_pre_hashed(state->regs_table, hash, deref);
if (entry)
return entry->data;
unsigned array_size = 1;
nir_deref *tail = &deref->deref;
while (tail->child) {
if (tail->child->deref_type == nir_deref_type_array)
array_size *= glsl_get_length(tail->type);
tail = tail->child;
}
assert(glsl_type_is_vector(tail->type) || glsl_type_is_scalar(tail->type));
nir_register *reg = nir_local_reg_create(state->impl);
reg->num_components = glsl_get_vector_elements(tail->type);
reg->num_array_elems = array_size > 1 ? array_size : 0;
reg->bit_size = glsl_get_bit_size(glsl_get_base_type(tail->type));
_mesa_hash_table_insert_pre_hashed(state->regs_table, hash, deref, reg);
nir_array_add(&state->derefs_array, nir_deref_var *, deref);
return reg;
}
static nir_register *
get_reg_for_deref(nir_deref_instr *deref, struct locals_to_regs_state *state)
{
uint32_t hash = hash_deref(deref);
assert(nir_deref_instr_get_variable(deref)->constant_initializer == NULL);
struct hash_entry *entry =
_mesa_hash_table_search_pre_hashed(state->regs_table, hash, deref);
if (entry)
return entry->data;
unsigned array_size = 1;
for (nir_deref_instr *d = deref; d; d = nir_deref_instr_parent(d)) {
if (d->deref_type == nir_deref_type_array)
array_size *= glsl_get_length(nir_deref_instr_parent(d)->type);
}
assert(glsl_type_is_vector_or_scalar(deref->type));
nir_register *reg = nir_local_reg_create(state->builder.impl);
reg->num_components = glsl_get_vector_elements(deref->type);
reg->num_array_elems = array_size > 1 ? array_size : 0;
reg->bit_size = glsl_get_bit_size(deref->type);
_mesa_hash_table_insert_pre_hashed(state->regs_table, hash, deref, reg);
return reg;
}
void
glsl_get_natural_size_align_bytes(const struct glsl_type *type,
unsigned *size, unsigned *align)
{
switch (type->base_type) {
case GLSL_TYPE_UINT8:
case GLSL_TYPE_INT8:
case GLSL_TYPE_UINT16:
case GLSL_TYPE_INT16:
case GLSL_TYPE_FLOAT16:
case GLSL_TYPE_UINT:
case GLSL_TYPE_INT:
case GLSL_TYPE_FLOAT:
case GLSL_TYPE_BOOL:
case GLSL_TYPE_DOUBLE:
case GLSL_TYPE_UINT64:
case GLSL_TYPE_INT64: {
unsigned N = glsl_get_bit_size(type) / 8;
*size = N * type->components();
*align = N;
break;
}
case GLSL_TYPE_ARRAY: {
unsigned elem_size, elem_align;
glsl_get_natural_size_align_bytes(type->fields.array,
&elem_size, &elem_align);
*align = elem_align;
*size = type->length * ALIGN_POT(elem_size, elem_align);
break;
}
case GLSL_TYPE_STRUCT:
*size = 0;
*align = 0;
for (unsigned i = 0; i < type->length; i++) {
unsigned elem_size, elem_align;
glsl_get_natural_size_align_bytes(type->fields.structure[i].type,
&elem_size, &elem_align);
*align = MAX2(*align, elem_align);
*size = ALIGN_POT(*size, elem_align) + elem_size;
}
break;
case GLSL_TYPE_SAMPLER:
case GLSL_TYPE_ATOMIC_UINT:
case GLSL_TYPE_SUBROUTINE:
case GLSL_TYPE_IMAGE:
case GLSL_TYPE_VOID:
case GLSL_TYPE_ERROR:
case GLSL_TYPE_INTERFACE:
case GLSL_TYPE_FUNCTION:
unreachable("type does not have a natural size");
}
}
static void
handle_glsl450_interpolation(struct vtn_builder *b, enum GLSLstd450 opcode,
const uint32_t *w, unsigned count)
{
const struct glsl_type *dest_type =
vtn_value(b, w[1], vtn_value_type_type)->type->type;
struct vtn_value *val = vtn_push_value(b, w[2], vtn_value_type_ssa);
val->ssa = vtn_create_ssa_value(b, dest_type);
nir_intrinsic_op op;
switch (opcode) {
case GLSLstd450InterpolateAtCentroid:
op = nir_intrinsic_interp_var_at_centroid;
break;
case GLSLstd450InterpolateAtSample:
op = nir_intrinsic_interp_var_at_sample;
break;
case GLSLstd450InterpolateAtOffset:
op = nir_intrinsic_interp_var_at_offset;
break;
default:
unreachable("Invalid opcode");
}
nir_intrinsic_instr *intrin = nir_intrinsic_instr_create(b->nb.shader, op);
nir_deref_var *deref = vtn_nir_deref(b, w[5]);
intrin->variables[0] =
nir_deref_as_var(nir_copy_deref(intrin, &deref->deref));
switch (opcode) {
case GLSLstd450InterpolateAtCentroid:
break;
case GLSLstd450InterpolateAtSample:
case GLSLstd450InterpolateAtOffset:
intrin->src[0] = nir_src_for_ssa(vtn_ssa_value(b, w[6])->def);
break;
default:
unreachable("Invalid opcode");
}
intrin->num_components = glsl_get_vector_elements(dest_type);
nir_ssa_dest_init(&intrin->instr, &intrin->dest,
glsl_get_vector_elements(dest_type),
glsl_get_bit_size(dest_type), NULL);
val->ssa->def = &intrin->dest.ssa;
nir_builder_instr_insert(&b->nb, &intrin->instr);
}
static void
_vtn_load_store_tail(struct vtn_builder *b, nir_intrinsic_op op, bool load,
nir_ssa_def *index, nir_ssa_def *offset,
struct vtn_ssa_value **inout, const struct glsl_type *type)
{
nir_intrinsic_instr *instr = nir_intrinsic_instr_create(b->nb.shader, op);
instr->num_components = glsl_get_vector_elements(type);
int src = 0;
if (!load) {
nir_intrinsic_set_write_mask(instr, (1 << instr->num_components) - 1);
instr->src[src++] = nir_src_for_ssa((*inout)->def);
}
/* We set the base and size for push constant load to the entire push
* constant block for now.
*/
if (op == nir_intrinsic_load_push_constant) {
nir_intrinsic_set_base(instr, 0);
nir_intrinsic_set_range(instr, 128);
}
if (index)
instr->src[src++] = nir_src_for_ssa(index);
instr->src[src++] = nir_src_for_ssa(offset);
if (load) {
nir_ssa_dest_init(&instr->instr, &instr->dest,
instr->num_components,
glsl_get_bit_size(glsl_get_base_type(type)), NULL);
(*inout)->def = &instr->dest.ssa;
}
nir_builder_instr_insert(&b->nb, &instr->instr);
if (load && glsl_get_base_type(type) == GLSL_TYPE_BOOL)
(*inout)->def = nir_ine(&b->nb, (*inout)->def, nir_imm_int(&b->nb, 0));
}
static void
validate_intrinsic_instr(nir_intrinsic_instr *instr, validate_state *state)
{
unsigned bit_size = 0;
if (instr->intrinsic == nir_intrinsic_load_var ||
instr->intrinsic == nir_intrinsic_store_var) {
const struct glsl_type *type =
nir_deref_tail(&instr->variables[0]->deref)->type;
bit_size = glsl_get_bit_size(type);
}
unsigned num_srcs = nir_intrinsic_infos[instr->intrinsic].num_srcs;
for (unsigned i = 0; i < num_srcs; i++) {
unsigned components_read =
nir_intrinsic_infos[instr->intrinsic].src_components[i];
if (components_read == 0)
components_read = instr->num_components;
validate_assert(state, components_read > 0);
validate_src(&instr->src[i], state, bit_size, components_read);
}
unsigned num_vars = nir_intrinsic_infos[instr->intrinsic].num_variables;
for (unsigned i = 0; i < num_vars; i++) {
validate_deref_var(instr, instr->variables[i], state);
}
if (nir_intrinsic_infos[instr->intrinsic].has_dest) {
unsigned components_written =
nir_intrinsic_infos[instr->intrinsic].dest_components;
if (components_written == 0)
components_written = instr->num_components;
validate_assert(state, components_written > 0);
validate_dest(&instr->dest, state, bit_size, components_written);
}
switch (instr->intrinsic) {
case nir_intrinsic_load_var: {
const struct glsl_type *type =
nir_deref_tail(&instr->variables[0]->deref)->type;
validate_assert(state, glsl_type_is_vector_or_scalar(type) ||
(instr->variables[0]->var->data.mode == nir_var_uniform &&
glsl_get_base_type(type) == GLSL_TYPE_SUBROUTINE));
validate_assert(state, instr->num_components == glsl_get_vector_elements(type));
break;
}
case nir_intrinsic_store_var: {
const struct glsl_type *type =
nir_deref_tail(&instr->variables[0]->deref)->type;
validate_assert(state, glsl_type_is_vector_or_scalar(type) ||
(instr->variables[0]->var->data.mode == nir_var_uniform &&
glsl_get_base_type(type) == GLSL_TYPE_SUBROUTINE));
validate_assert(state, instr->num_components == glsl_get_vector_elements(type));
validate_assert(state, instr->variables[0]->var->data.mode != nir_var_shader_in &&
instr->variables[0]->var->data.mode != nir_var_uniform &&
instr->variables[0]->var->data.mode != nir_var_shader_storage);
validate_assert(state, (nir_intrinsic_write_mask(instr) & ~((1 << instr->num_components) - 1)) == 0);
break;
}
case nir_intrinsic_copy_var:
validate_assert(state, nir_deref_tail(&instr->variables[0]->deref)->type ==
nir_deref_tail(&instr->variables[1]->deref)->type);
validate_assert(state, instr->variables[0]->var->data.mode != nir_var_shader_in &&
instr->variables[0]->var->data.mode != nir_var_uniform &&
instr->variables[0]->var->data.mode != nir_var_shader_storage);
break;
default:
break;
}
}
static void
_vtn_local_load_store(struct vtn_builder *b, bool load, nir_deref_var *deref,
nir_deref *tail, struct vtn_ssa_value *inout)
{
/* The deref tail may contain a deref to select a component of a vector (in
* other words, it might not be an actual tail) so we have to save it away
* here since we overwrite it later.
*/
nir_deref *old_child = tail->child;
if (glsl_type_is_vector_or_scalar(tail->type)) {
/* Terminate the deref chain in case there is one more link to pick
* off a component of the vector.
*/
tail->child = NULL;
nir_intrinsic_op op = load ? nir_intrinsic_load_var :
nir_intrinsic_store_var;
nir_intrinsic_instr *intrin = nir_intrinsic_instr_create(b->shader, op);
intrin->variables[0] =
nir_deref_as_var(nir_copy_deref(intrin, &deref->deref));
intrin->num_components = glsl_get_vector_elements(tail->type);
if (load) {
nir_ssa_dest_init(&intrin->instr, &intrin->dest,
intrin->num_components,
glsl_get_bit_size(glsl_get_base_type(tail->type)),
NULL);
inout->def = &intrin->dest.ssa;
} else {
nir_intrinsic_set_write_mask(intrin, (1 << intrin->num_components) - 1);
intrin->src[0] = nir_src_for_ssa(inout->def);
}
nir_builder_instr_insert(&b->nb, &intrin->instr);
} else if (glsl_get_base_type(tail->type) == GLSL_TYPE_ARRAY ||
glsl_type_is_matrix(tail->type)) {
unsigned elems = glsl_get_length(tail->type);
nir_deref_array *deref_arr = nir_deref_array_create(b);
deref_arr->deref_array_type = nir_deref_array_type_direct;
deref_arr->deref.type = glsl_get_array_element(tail->type);
tail->child = &deref_arr->deref;
for (unsigned i = 0; i < elems; i++) {
deref_arr->base_offset = i;
_vtn_local_load_store(b, load, deref, tail->child, inout->elems[i]);
}
} else {
assert(glsl_get_base_type(tail->type) == GLSL_TYPE_STRUCT);
unsigned elems = glsl_get_length(tail->type);
nir_deref_struct *deref_struct = nir_deref_struct_create(b, 0);
tail->child = &deref_struct->deref;
for (unsigned i = 0; i < elems; i++) {
deref_struct->index = i;
deref_struct->deref.type = glsl_get_struct_field(tail->type, i);
_vtn_local_load_store(b, load, deref, tail->child, inout->elems[i]);
}
}
tail->child = old_child;
}
