#include "nir.h"
#include "nir_builder.h"
+#include "nir_deref.h"
struct lower_io_state {
+ void *dead_ctx;
nir_builder builder;
- void *mem_ctx;
int (*type_size)(const struct glsl_type *type);
nir_variable_mode modes;
+ nir_lower_io_options options;
};
void
nir_assign_var_locations(struct exec_list *var_list, unsigned *size,
- unsigned base_offset,
int (*type_size)(const struct glsl_type *))
{
unsigned location = 0;
- /* There are 32 regular and 32 patch varyings allowed */
- int locations[64][2];
- for (unsigned i = 0; i < 64; i++) {
- for (unsigned j = 0; j < 2; j++)
- locations[i][j] = -1;
- }
-
nir_foreach_variable(var, var_list) {
/*
- * UBO's have their own address spaces, so don't count them towards the
+ * UBOs have their own address spaces, so don't count them towards the
* number of global uniforms
*/
if ((var->data.mode == nir_var_uniform || var->data.mode == nir_var_shader_storage) &&
var->interface_type != NULL)
continue;
- /* Make sure we give the same location to varyings packed with
- * ARB_enhanced_layouts.
- */
- int idx = var->data.location - base_offset;
- if (base_offset && idx >= 0) {
- assert(idx < ARRAY_SIZE(locations));
-
- if (locations[idx][var->data.index] == -1) {
- var->data.driver_location = location;
- locations[idx][var->data.index] = location;
- location += type_size(var->type);
- } else {
- var->data.driver_location = locations[idx][var->data.index];
- }
- } else {
- var->data.driver_location = location;
- location += type_size(var->type);
- }
+ var->data.driver_location = location;
+ location += type_size(var->type);
}
*size = location;
}
/**
- * Returns true if we're processing a stage whose inputs are arrays indexed
- * by a vertex number (such as geometry shader inputs).
+ * Return true if the given variable is a per-vertex input/output array.
+ * (such as geometry shader inputs).
*/
-static bool
-is_per_vertex_input(struct lower_io_state *state, nir_variable *var)
+bool
+nir_is_per_vertex_io(const nir_variable *var, gl_shader_stage stage)
{
- gl_shader_stage stage = state->builder.shader->stage;
+ if (var->data.patch || !glsl_type_is_array(var->type))
+ return false;
- return var->data.mode == nir_var_shader_in && !var->data.patch &&
- (stage == MESA_SHADER_TESS_CTRL ||
- stage == MESA_SHADER_TESS_EVAL ||
- stage == MESA_SHADER_GEOMETRY);
-}
+ if (var->data.mode == nir_var_shader_in)
+ return stage == MESA_SHADER_GEOMETRY ||
+ stage == MESA_SHADER_TESS_CTRL ||
+ stage == MESA_SHADER_TESS_EVAL;
-static bool
-is_per_vertex_output(struct lower_io_state *state, nir_variable *var)
-{
- gl_shader_stage stage = state->builder.shader->stage;
- return var->data.mode == nir_var_shader_out && !var->data.patch &&
- stage == MESA_SHADER_TESS_CTRL;
+ if (var->data.mode == nir_var_shader_out)
+ return stage == MESA_SHADER_TESS_CTRL;
+
+ return false;
}
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 *))
+ 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;
- assert(tail->deref_type == nir_deref_type_array);
- nir_deref_array *deref_array = nir_deref_as_array(tail);
+ assert((*p)->deref_type == nir_deref_type_array);
+ *vertex_index = nir_ssa_for_src(b, (*p)->arr.index, 1);
+ p++;
+ }
- 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;
+ 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. */
+ nir_const_value *const_index = nir_src_as_const_value((*p)->arr.index);
+ assert(const_index);
+
+ 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_op
-load_op(nir_variable_mode mode, bool per_vertex)
+static nir_intrinsic_instr *
+lower_load(nir_intrinsic_instr *intrin, struct lower_io_state *state,
+ nir_ssa_def *vertex_index, nir_variable *var, nir_ssa_def *offset,
+ unsigned component)
{
+ const nir_shader *nir = state->builder.shader;
+ nir_variable_mode mode = var->data.mode;
+ nir_ssa_def *barycentric = NULL;
+
nir_intrinsic_op op;
switch (mode) {
case nir_var_shader_in:
- op = per_vertex ? nir_intrinsic_load_per_vertex_input :
- nir_intrinsic_load_input;
+ if (nir->info.stage == MESA_SHADER_FRAGMENT &&
+ nir->options->use_interpolated_input_intrinsics &&
+ var->data.interpolation != INTERP_MODE_FLAT) {
+ assert(vertex_index == NULL);
+
+ nir_intrinsic_op bary_op;
+ if (var->data.sample ||
+ (state->options & nir_lower_io_force_sample_interpolation))
+ bary_op = nir_intrinsic_load_barycentric_sample;
+ else if (var->data.centroid)
+ bary_op = nir_intrinsic_load_barycentric_centroid;
+ else
+ bary_op = nir_intrinsic_load_barycentric_pixel;
+
+ barycentric = nir_load_barycentric(&state->builder, bary_op,
+ var->data.interpolation);
+ op = nir_intrinsic_load_interpolated_input;
+ } else {
+ op = vertex_index ? nir_intrinsic_load_per_vertex_input :
+ nir_intrinsic_load_input;
+ }
break;
case nir_var_shader_out:
- op = per_vertex ? nir_intrinsic_load_per_vertex_output :
- nir_intrinsic_load_output;
+ op = vertex_index ? nir_intrinsic_load_per_vertex_output :
+ nir_intrinsic_load_output;
break;
case nir_var_uniform:
op = nir_intrinsic_load_uniform;
default:
unreachable("Unknown variable mode");
}
- return op;
+
+ nir_intrinsic_instr *load =
+ nir_intrinsic_instr_create(state->builder.shader, op);
+ load->num_components = intrin->num_components;
+
+ nir_intrinsic_set_base(load, var->data.driver_location);
+ if (mode == nir_var_shader_in || mode == nir_var_shader_out)
+ nir_intrinsic_set_component(load, component);
+
+ if (load->intrinsic == nir_intrinsic_load_uniform)
+ nir_intrinsic_set_range(load, state->type_size(var->type));
+
+ if (vertex_index) {
+ load->src[0] = nir_src_for_ssa(vertex_index);
+ load->src[1] = nir_src_for_ssa(offset);
+ } else if (barycentric) {
+ load->src[0] = nir_src_for_ssa(barycentric);
+ load->src[1] = nir_src_for_ssa(offset);
+ } else {
+ load->src[0] = nir_src_for_ssa(offset);
+ }
+
+ return load;
}
-static nir_intrinsic_op
-store_op(struct lower_io_state *state,
- nir_variable_mode mode, bool per_vertex)
+static nir_intrinsic_instr *
+lower_store(nir_intrinsic_instr *intrin, struct lower_io_state *state,
+ nir_ssa_def *vertex_index, nir_variable *var, nir_ssa_def *offset,
+ unsigned component)
{
+ nir_variable_mode mode = var->data.mode;
+
nir_intrinsic_op op;
- switch (mode) {
- case nir_var_shader_in:
- case nir_var_shader_out:
- op = per_vertex ? nir_intrinsic_store_per_vertex_output :
- nir_intrinsic_store_output;
- break;
- case nir_var_shared:
+ if (mode == nir_var_shared) {
op = nir_intrinsic_store_shared;
- break;
- default:
- unreachable("Unknown variable mode");
+ } else {
+ assert(mode == nir_var_shader_out);
+ op = vertex_index ? nir_intrinsic_store_per_vertex_output :
+ nir_intrinsic_store_output;
}
- return op;
+
+ nir_intrinsic_instr *store =
+ nir_intrinsic_instr_create(state->builder.shader, op);
+ store->num_components = intrin->num_components;
+
+ nir_src_copy(&store->src[0], &intrin->src[1], store);
+
+ nir_intrinsic_set_base(store, var->data.driver_location);
+
+ if (mode == nir_var_shader_out)
+ nir_intrinsic_set_component(store, component);
+
+ nir_intrinsic_set_write_mask(store, nir_intrinsic_write_mask(intrin));
+
+ if (vertex_index)
+ store->src[1] = nir_src_for_ssa(vertex_index);
+
+ store->src[vertex_index ? 2 : 1] = nir_src_for_ssa(offset);
+
+ return store;
}
-static nir_intrinsic_op
-atomic_op(nir_intrinsic_op opcode)
+static nir_intrinsic_instr *
+lower_atomic(nir_intrinsic_instr *intrin, struct lower_io_state *state,
+ nir_variable *var, nir_ssa_def *offset)
{
- switch (opcode) {
-#define OP(O) case nir_intrinsic_var_##O: return nir_intrinsic_shared_##O;
+ assert(var->data.mode == nir_var_shared);
+
+ nir_intrinsic_op op;
+ switch (intrin->intrinsic) {
+#define OP(O) case nir_intrinsic_deref_##O: op = nir_intrinsic_shared_##O; break;
OP(atomic_exchange)
OP(atomic_comp_swap)
OP(atomic_add)
default:
unreachable("Invalid atomic");
}
+
+ nir_intrinsic_instr *atomic =
+ nir_intrinsic_instr_create(state->builder.shader, op);
+
+ nir_intrinsic_set_base(atomic, var->data.driver_location);
+
+ atomic->src[0] = nir_src_for_ssa(offset);
+ 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_variable *var, nir_ssa_def *offset, unsigned component)
+{
+ 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, var, offset, component);
+
+ nir_intrinsic_op bary_op;
+ switch (intrin->intrinsic) {
+ 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_deref_at_sample:
+ bary_op = nir_intrinsic_load_barycentric_at_sample;
+ break;
+ case nir_intrinsic_interp_deref_at_offset:
+ bary_op = nir_intrinsic_load_barycentric_at_offset;
+ break;
+ default:
+ unreachable("Bogus interpolateAt() intrinsic.");
+ }
+
+ nir_intrinsic_instr *bary_setup =
+ nir_intrinsic_instr_create(state->builder.shader, bary_op);
+
+ 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_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 *load =
+ nir_intrinsic_instr_create(state->builder.shader,
+ nir_intrinsic_load_interpolated_input);
+ load->num_components = intrin->num_components;
+
+ nir_intrinsic_set_base(load, var->data.driver_location);
+ nir_intrinsic_set_component(load, component);
+
+ load->src[0] = nir_src_for_ssa(&bary_setup->dest.ssa);
+ load->src[1] = nir_src_for_ssa(offset);
+
+ return load;
}
static bool
struct lower_io_state *state)
{
nir_builder *b = &state->builder;
+ const nir_shader_compiler_options *options = b->shader->options;
+ bool progress = false;
nir_foreach_instr_safe(instr, block) {
if (instr->type != nir_instr_type_intrinsic)
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_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;
default:
/* We can't lower the io for this nir instrinsic, so skip it */
continue;
}
- nir_variable_mode mode = intrin->variables[0]->var->data.mode;
+ 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)
continue;
b->cursor = nir_before_instr(instr);
- switch (intrin->intrinsic) {
- case nir_intrinsic_load_var: {
- bool per_vertex =
- is_per_vertex_input(state, intrin->variables[0]->var) ||
- is_per_vertex_output(state, intrin->variables[0]->var);
-
- nir_ssa_def *offset;
- nir_ssa_def *vertex_index;
-
- offset = get_io_offset(b, intrin->variables[0],
- per_vertex ? &vertex_index : NULL,
- state->type_size);
-
- nir_intrinsic_instr *load =
- nir_intrinsic_instr_create(state->mem_ctx,
- load_op(mode, per_vertex));
- load->num_components = intrin->num_components;
-
- nir_intrinsic_set_base(load,
- intrin->variables[0]->var->data.driver_location);
- if (mode == nir_var_shader_in || mode == nir_var_shader_out) {
- nir_intrinsic_set_component(load,
- intrin->variables[0]->var->data.location_frac);
- }
-
- if (load->intrinsic == nir_intrinsic_load_uniform) {
- nir_intrinsic_set_range(load,
- state->type_size(intrin->variables[0]->var->type));
- }
+ const bool per_vertex = nir_is_per_vertex_io(var, b->shader->info.stage);
- if (per_vertex)
- load->src[0] = nir_src_for_ssa(vertex_index);
+ nir_ssa_def *offset;
+ nir_ssa_def *vertex_index = NULL;
+ unsigned component_offset = var->data.location_frac;
- load->src[per_vertex ? 1 : 0] = nir_src_for_ssa(offset);
+ offset = get_io_offset(b, deref, per_vertex ? &vertex_index : NULL,
+ state->type_size, &component_offset);
- if (intrin->dest.is_ssa) {
- nir_ssa_dest_init(&load->instr, &load->dest,
- intrin->num_components,
- intrin->dest.ssa.bit_size, NULL);
- nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
- nir_src_for_ssa(&load->dest.ssa));
- } else {
- nir_dest_copy(&load->dest, &intrin->dest, state->mem_ctx);
- }
+ nir_intrinsic_instr *replacement;
- nir_instr_insert_before(&intrin->instr, &load->instr);
- nir_instr_remove(&intrin->instr);
+ switch (intrin->intrinsic) {
+ case nir_intrinsic_load_deref:
+ replacement = lower_load(intrin, state, vertex_index, var, offset,
+ component_offset);
break;
- }
-
- case nir_intrinsic_store_var: {
- assert(mode == nir_var_shader_out || mode == nir_var_shared);
-
- nir_ssa_def *offset;
- nir_ssa_def *vertex_index;
-
- bool per_vertex =
- is_per_vertex_output(state, intrin->variables[0]->var);
-
- offset = get_io_offset(b, intrin->variables[0],
- per_vertex ? &vertex_index : NULL,
- state->type_size);
-
- nir_intrinsic_instr *store =
- nir_intrinsic_instr_create(state->mem_ctx,
- store_op(state, mode, per_vertex));
- store->num_components = intrin->num_components;
- nir_src_copy(&store->src[0], &intrin->src[0], store);
-
- nir_intrinsic_set_base(store,
- intrin->variables[0]->var->data.driver_location);
- if (mode == nir_var_shader_out) {
- nir_intrinsic_set_component(store,
- intrin->variables[0]->var->data.location_frac);
- }
- nir_intrinsic_set_write_mask(store, nir_intrinsic_write_mask(intrin));
-
- if (per_vertex)
- store->src[1] = nir_src_for_ssa(vertex_index);
-
- store->src[per_vertex ? 2 : 1] = nir_src_for_ssa(offset);
-
- nir_instr_insert_before(&intrin->instr, &store->instr);
- nir_instr_remove(&intrin->instr);
+ 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: {
- assert(mode == nir_var_shared);
- nir_ssa_def *offset;
-
- offset = get_io_offset(b, intrin->variables[0],
- NULL, state->type_size);
-
- nir_intrinsic_instr *atomic =
- nir_intrinsic_instr_create(state->mem_ctx,
- atomic_op(intrin->intrinsic));
-
- atomic->src[0] = nir_src_for_ssa(offset);
+ 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, var, offset);
+ break;
- atomic->const_index[0] =
- intrin->variables[0]->var->data.driver_location;
+ 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, var, offset,
+ component_offset);
+ break;
- for (unsigned i = 0;
- i < nir_op_infos[intrin->intrinsic].num_inputs;
- i++) {
- nir_src_copy(&atomic->src[i+1], &intrin->src[i], atomic);
- }
+ default:
+ continue;
+ }
+ if (nir_intrinsic_infos[intrin->intrinsic].has_dest) {
if (intrin->dest.is_ssa) {
- nir_ssa_dest_init(&atomic->instr, &atomic->dest,
+ nir_ssa_dest_init(&replacement->instr, &replacement->dest,
intrin->dest.ssa.num_components,
intrin->dest.ssa.bit_size, NULL);
nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
- nir_src_for_ssa(&atomic->dest.ssa));
+ nir_src_for_ssa(&replacement->dest.ssa));
} else {
- nir_dest_copy(&atomic->dest, &intrin->dest, state->mem_ctx);
+ nir_dest_copy(&replacement->dest, &intrin->dest, &intrin->instr);
}
-
- nir_instr_insert_before(&intrin->instr, &atomic->instr);
- nir_instr_remove(&intrin->instr);
- break;
}
- default:
- break;
- }
+ nir_instr_insert_before(&intrin->instr, &replacement->instr);
+ nir_instr_remove(&intrin->instr);
+ progress = true;
}
- return true;
+ return progress;
}
-static void
+static bool
nir_lower_io_impl(nir_function_impl *impl,
nir_variable_mode modes,
- int (*type_size)(const struct glsl_type *))
+ int (*type_size)(const struct glsl_type *),
+ nir_lower_io_options options)
{
struct lower_io_state state;
+ bool progress = false;
nir_builder_init(&state.builder, impl);
- state.mem_ctx = ralloc_parent(impl);
+ state.dead_ctx = ralloc_context(NULL);
state.modes = modes;
state.type_size = type_size;
+ state.options = options;
nir_foreach_block(block, impl) {
- nir_lower_io_block(block, &state);
+ 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;
}
-void
+bool
nir_lower_io(nir_shader *shader, nir_variable_mode modes,
- int (*type_size)(const struct glsl_type *))
+ int (*type_size)(const struct glsl_type *),
+ nir_lower_io_options options)
{
+ bool progress = false;
+
+ 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)
- nir_lower_io_impl(function->impl, modes, type_size);
+ if (function->impl) {
+ progress |= nir_lower_io_impl(function->impl, modes,
+ type_size, options);
+ }
}
+
+ return progress;
}
/**
- * Return the offset soruce for a load/store intrinsic.
+ * Return the offset source for a load/store intrinsic.
*/
nir_src *
nir_get_io_offset_src(nir_intrinsic_instr *instr)
case nir_intrinsic_load_ssbo:
case nir_intrinsic_load_per_vertex_input:
case nir_intrinsic_load_per_vertex_output:
+ case nir_intrinsic_load_interpolated_input:
case nir_intrinsic_store_output:
return &instr->src[1];
case nir_intrinsic_store_ssbo:
projects / mesa.git / blobdiff