#include "nir.h"
#include "nir_builder.h"
+#include "nir_deref.h"
struct lower_io_state {
+ void *dead_ctx;
nir_builder builder;
int (*type_size)(const struct glsl_type *type);
nir_variable_mode modes;
}
static nir_ssa_def *
-get_io_offset(nir_builder *b, nir_deref_var *deref,
+get_io_offset(nir_builder *b, nir_deref_instr *deref,
nir_ssa_def **vertex_index,
int (*type_size)(const struct glsl_type *),
unsigned *component)
{
- nir_deref *tail = &deref->deref;
+ nir_deref_path path;
+ nir_deref_path_init(&path, deref, NULL);
+
+ 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), keep the
* outermost array index separate. Process the rest normally.
*/
if (vertex_index != NULL) {
- tail = tail->child;
- nir_deref_array *deref_array = nir_deref_as_array(tail);
-
- nir_ssa_def *vtx = nir_imm_int(b, deref_array->base_offset);
- if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
- vtx = nir_iadd(b, vtx, nir_ssa_for_src(b, deref_array->indirect, 1));
- }
- *vertex_index = vtx;
+ assert((*p)->deref_type == nir_deref_type_array);
+ *vertex_index = nir_ssa_for_src(b, (*p)->arr.index, 1);
+ p++;
}
- if (deref->var->data.compact) {
- assert(tail->child->deref_type == nir_deref_type_array);
- assert(glsl_type_is_scalar(glsl_without_array(deref->var->type)));
- nir_deref_array *deref_array = nir_deref_as_array(tail->child);
+ if (path.path[0]->var->data.compact) {
+ assert((*p)->deref_type == nir_deref_type_array);
+ assert(glsl_type_is_scalar((*p)->type));
+
/* We always lower indirect dereferences for "compact" array vars. */
- assert(deref_array->deref_array_type == nir_deref_array_type_direct);
+ nir_const_value *const_index = nir_src_as_const_value((*p)->arr.index);
+ assert(const_index);
- const unsigned total_offset = *component + deref_array->base_offset;
+ const unsigned total_offset = *component + const_index->u32[0];
const unsigned slot_offset = total_offset / 4;
*component = total_offset % 4;
return nir_imm_int(b, type_size(glsl_vec4_type()) * slot_offset);
/* Just emit code and let constant-folding go to town */
nir_ssa_def *offset = nir_imm_int(b, 0);
- while (tail->child != NULL) {
- const struct glsl_type *parent_type = tail->type;
- tail = tail->child;
+ for (; *p; p++) {
+ if ((*p)->deref_type == nir_deref_type_array) {
+ unsigned size = type_size((*p)->type);
- if (tail->deref_type == nir_deref_type_array) {
- nir_deref_array *deref_array = nir_deref_as_array(tail);
- unsigned size = type_size(tail->type);
+ nir_ssa_def *mul =
+ nir_imul(b, nir_imm_int(b, size),
+ nir_ssa_for_src(b, (*p)->arr.index, 1));
- offset = nir_iadd(b, offset,
- nir_imm_int(b, size * deref_array->base_offset));
-
- if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
- nir_ssa_def *mul =
- nir_imul(b, nir_imm_int(b, size),
- nir_ssa_for_src(b, deref_array->indirect, 1));
-
- offset = nir_iadd(b, offset, mul);
- }
- } else if (tail->deref_type == nir_deref_type_struct) {
- nir_deref_struct *deref_struct = nir_deref_as_struct(tail);
+ offset = nir_iadd(b, offset, mul);
+ } else if ((*p)->deref_type == nir_deref_type_struct) {
+ /* p starts at path[1], so this is safe */
+ nir_deref_instr *parent = *(p - 1);
unsigned field_offset = 0;
- for (unsigned i = 0; i < deref_struct->index; i++) {
- field_offset += type_size(glsl_get_struct_field(parent_type, i));
+ for (unsigned i = 0; i < (*p)->strct.index; i++) {
+ field_offset += type_size(glsl_get_struct_field(parent->type, i));
}
offset = nir_iadd(b, offset, nir_imm_int(b, field_offset));
+ } else {
+ unreachable("Unsupported deref type");
}
}
+ nir_deref_path_finish(&path);
+
return offset;
}
static nir_intrinsic_instr *
lower_load(nir_intrinsic_instr *intrin, struct lower_io_state *state,
- nir_ssa_def *vertex_index, nir_ssa_def *offset,
+ nir_ssa_def *vertex_index, nir_variable *var, nir_ssa_def *offset,
unsigned component)
{
const nir_shader *nir = state->builder.shader;
- nir_variable *var = intrin->variables[0]->var;
nir_variable_mode mode = var->data.mode;
nir_ssa_def *barycentric = NULL;
static nir_intrinsic_instr *
lower_store(nir_intrinsic_instr *intrin, struct lower_io_state *state,
- nir_ssa_def *vertex_index, nir_ssa_def *offset,
+ nir_ssa_def *vertex_index, nir_variable *var, nir_ssa_def *offset,
unsigned component)
{
- nir_variable *var = intrin->variables[0]->var;
nir_variable_mode mode = var->data.mode;
nir_intrinsic_op op;
nir_intrinsic_instr_create(state->builder.shader, op);
store->num_components = intrin->num_components;
- nir_src_copy(&store->src[0], &intrin->src[0], store);
+ nir_src_copy(&store->src[0], &intrin->src[1], store);
nir_intrinsic_set_base(store, var->data.driver_location);
static nir_intrinsic_instr *
lower_atomic(nir_intrinsic_instr *intrin, struct lower_io_state *state,
- nir_ssa_def *offset)
+ nir_variable *var, nir_ssa_def *offset)
{
- nir_variable *var = intrin->variables[0]->var;
-
assert(var->data.mode == nir_var_shared);
nir_intrinsic_op op;
switch (intrin->intrinsic) {
-#define OP(O) case nir_intrinsic_var_##O: op = nir_intrinsic_shared_##O; break;
+#define OP(O) case nir_intrinsic_deref_##O: op = nir_intrinsic_shared_##O; break;
OP(atomic_exchange)
OP(atomic_comp_swap)
OP(atomic_add)
nir_intrinsic_set_base(atomic, var->data.driver_location);
atomic->src[0] = nir_src_for_ssa(offset);
- for (unsigned i = 0; i < nir_intrinsic_infos[intrin->intrinsic].num_srcs; i++) {
- nir_src_copy(&atomic->src[i+1], &intrin->src[i], atomic);
+ assert(nir_intrinsic_infos[intrin->intrinsic].num_srcs ==
+ nir_intrinsic_infos[op].num_srcs);
+ for (unsigned i = 1; i < nir_intrinsic_infos[op].num_srcs; i++) {
+ nir_src_copy(&atomic->src[i], &intrin->src[i], atomic);
}
return atomic;
static nir_intrinsic_instr *
lower_interpolate_at(nir_intrinsic_instr *intrin, struct lower_io_state *state,
- nir_ssa_def *offset, unsigned component)
+ nir_variable *var, nir_ssa_def *offset, unsigned component)
{
- nir_variable *var = intrin->variables[0]->var;
-
assert(var->data.mode == nir_var_shader_in);
/* Ignore interpolateAt() for flat variables - flat is flat. */
if (var->data.interpolation == INTERP_MODE_FLAT)
- return lower_load(intrin, state, NULL, offset, component);
+ return lower_load(intrin, state, NULL, var, offset, component);
nir_intrinsic_op bary_op;
switch (intrin->intrinsic) {
- case nir_intrinsic_interp_var_at_centroid:
+ case nir_intrinsic_interp_deref_at_centroid:
bary_op = (state->options & nir_lower_io_force_sample_interpolation) ?
nir_intrinsic_load_barycentric_sample :
nir_intrinsic_load_barycentric_centroid;
break;
- case nir_intrinsic_interp_var_at_sample:
+ case nir_intrinsic_interp_deref_at_sample:
bary_op = nir_intrinsic_load_barycentric_at_sample;
break;
- case nir_intrinsic_interp_var_at_offset:
+ case nir_intrinsic_interp_deref_at_offset:
bary_op = nir_intrinsic_load_barycentric_at_offset;
break;
default:
nir_ssa_dest_init(&bary_setup->instr, &bary_setup->dest, 2, 32, NULL);
nir_intrinsic_set_interp_mode(bary_setup, var->data.interpolation);
- if (intrin->intrinsic != nir_intrinsic_interp_var_at_centroid)
- nir_src_copy(&bary_setup->src[0], &intrin->src[0], bary_setup);
+ if (intrin->intrinsic == nir_intrinsic_interp_deref_at_sample ||
+ intrin->intrinsic == nir_intrinsic_interp_deref_at_offset)
+ nir_src_copy(&bary_setup->src[0], &intrin->src[1], bary_setup);
nir_builder_instr_insert(&state->builder, &bary_setup->instr);
nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
switch (intrin->intrinsic) {
- case nir_intrinsic_load_var:
- case nir_intrinsic_store_var:
- case nir_intrinsic_var_atomic_add:
- case nir_intrinsic_var_atomic_imin:
- case nir_intrinsic_var_atomic_umin:
- case nir_intrinsic_var_atomic_imax:
- case nir_intrinsic_var_atomic_umax:
- case nir_intrinsic_var_atomic_and:
- case nir_intrinsic_var_atomic_or:
- case nir_intrinsic_var_atomic_xor:
- case nir_intrinsic_var_atomic_exchange:
- case nir_intrinsic_var_atomic_comp_swap:
+ case nir_intrinsic_load_deref:
+ case nir_intrinsic_store_deref:
+ case nir_intrinsic_deref_atomic_add:
+ case nir_intrinsic_deref_atomic_imin:
+ case nir_intrinsic_deref_atomic_umin:
+ case nir_intrinsic_deref_atomic_imax:
+ case nir_intrinsic_deref_atomic_umax:
+ case nir_intrinsic_deref_atomic_and:
+ case nir_intrinsic_deref_atomic_or:
+ case nir_intrinsic_deref_atomic_xor:
+ case nir_intrinsic_deref_atomic_exchange:
+ case nir_intrinsic_deref_atomic_comp_swap:
/* We can lower the io for this nir instrinsic */
break;
- case nir_intrinsic_interp_var_at_centroid:
- case nir_intrinsic_interp_var_at_sample:
- case nir_intrinsic_interp_var_at_offset:
+ case nir_intrinsic_interp_deref_at_centroid:
+ case nir_intrinsic_interp_deref_at_sample:
+ case nir_intrinsic_interp_deref_at_offset:
/* We can optionally lower these to load_interpolated_input */
if (options->use_interpolated_input_intrinsics)
break;
continue;
}
- nir_variable *var = intrin->variables[0]->var;
+ nir_deref_instr *deref = nir_src_as_deref(intrin->src[0]);
+
+ nir_variable *var = nir_deref_instr_get_variable(deref);
nir_variable_mode mode = var->data.mode;
if ((state->modes & mode) == 0)
nir_ssa_def *vertex_index = NULL;
unsigned component_offset = var->data.location_frac;
- offset = get_io_offset(b, intrin->variables[0],
- per_vertex ? &vertex_index : NULL,
+ offset = get_io_offset(b, deref, per_vertex ? &vertex_index : NULL,
state->type_size, &component_offset);
nir_intrinsic_instr *replacement;
switch (intrin->intrinsic) {
- case nir_intrinsic_load_var:
- replacement = lower_load(intrin, state, vertex_index, offset,
+ case nir_intrinsic_load_deref:
+ replacement = lower_load(intrin, state, vertex_index, var, offset,
component_offset);
break;
- case nir_intrinsic_store_var:
- replacement = lower_store(intrin, state, vertex_index, offset,
+ case nir_intrinsic_store_deref:
+ replacement = lower_store(intrin, state, vertex_index, var, offset,
component_offset);
break;
- case nir_intrinsic_var_atomic_add:
- case nir_intrinsic_var_atomic_imin:
- case nir_intrinsic_var_atomic_umin:
- case nir_intrinsic_var_atomic_imax:
- case nir_intrinsic_var_atomic_umax:
- case nir_intrinsic_var_atomic_and:
- case nir_intrinsic_var_atomic_or:
- case nir_intrinsic_var_atomic_xor:
- case nir_intrinsic_var_atomic_exchange:
- case nir_intrinsic_var_atomic_comp_swap:
+ case nir_intrinsic_deref_atomic_add:
+ case nir_intrinsic_deref_atomic_imin:
+ case nir_intrinsic_deref_atomic_umin:
+ case nir_intrinsic_deref_atomic_imax:
+ case nir_intrinsic_deref_atomic_umax:
+ case nir_intrinsic_deref_atomic_and:
+ case nir_intrinsic_deref_atomic_or:
+ case nir_intrinsic_deref_atomic_xor:
+ case nir_intrinsic_deref_atomic_exchange:
+ case nir_intrinsic_deref_atomic_comp_swap:
assert(vertex_index == NULL);
- replacement = lower_atomic(intrin, state, offset);
+ replacement = lower_atomic(intrin, state, var, offset);
break;
- case nir_intrinsic_interp_var_at_centroid:
- case nir_intrinsic_interp_var_at_sample:
- case nir_intrinsic_interp_var_at_offset:
+ case nir_intrinsic_interp_deref_at_centroid:
+ case nir_intrinsic_interp_deref_at_sample:
+ case nir_intrinsic_interp_deref_at_offset:
assert(vertex_index == NULL);
- replacement = lower_interpolate_at(intrin, state, offset,
+ replacement = lower_interpolate_at(intrin, state, var, offset,
component_offset);
break;
bool progress = false;
nir_builder_init(&state.builder, impl);
+ state.dead_ctx = ralloc_context(NULL);
state.modes = modes;
state.type_size = type_size;
state.options = options;
progress |= nir_lower_io_block(block, &state);
}
+ ralloc_free(state.dead_ctx);
+
nir_metadata_preserve(impl, nir_metadata_block_index |
nir_metadata_dominance);
return progress;
{
bool progress = false;
- nir_assert_lowered_derefs(shader, nir_lower_load_store_derefs | nir_lower_interp_derefs | nir_lower_atomic_derefs);
+ nir_assert_unlowered_derefs(shader, nir_lower_load_store_derefs | nir_lower_interp_derefs | nir_lower_atomic_derefs);
nir_foreach_function(function, shader) {
if (function->impl) {
projects / mesa.git / blobdiff