case nir_intrinsic_load_base_vertex:
info->uses_basevertex = 1;
break;
+ case nir_intrinsic_load_draw_id:
+ info->uses_drawid = 1;
+ break;
case nir_intrinsic_load_primitive_id:
info->uses_primid = 1;
break;
case nir_intrinsic_load_tess_level_outer:
info->reads_tess_factors = true;
break;
- case nir_intrinsic_image_deref_load: {
- nir_variable *var = intrinsic_get_var(intr);
- if (var->data.bindless) {
- info->uses_bindless_images = true;
+ case nir_intrinsic_bindless_image_load:
+ info->uses_bindless_images = true;
- if (glsl_get_sampler_dim(var->type) == GLSL_SAMPLER_DIM_BUF)
- info->uses_bindless_buffer_load = true;
- else
- info->uses_bindless_image_load = true;
- }
+ if (nir_intrinsic_image_dim(intr) == GLSL_SAMPLER_DIM_BUF)
+ info->uses_bindless_buffer_load = true;
+ else
+ info->uses_bindless_image_load = true;
break;
- }
- case nir_intrinsic_image_deref_size:
- case nir_intrinsic_image_deref_samples: {
- nir_variable *var = intrinsic_get_var(intr);
- if (var->data.bindless)
- info->uses_bindless_images = true;
+ case nir_intrinsic_bindless_image_size:
+ case nir_intrinsic_bindless_image_samples:
+ info->uses_bindless_images = true;
break;
- }
- case nir_intrinsic_image_deref_store: {
- const nir_deref_instr *image_deref = nir_instr_as_deref(intr->src[0].ssa->parent_instr);
- nir_variable *var = intrinsic_get_var(intr);
- if (var->data.bindless) {
- info->uses_bindless_images = true;
+ case nir_intrinsic_bindless_image_store:
+ info->uses_bindless_images = true;
+
+ if (nir_intrinsic_image_dim(intr) == GLSL_SAMPLER_DIM_BUF)
+ info->uses_bindless_buffer_store = true;
+ else
+ info->uses_bindless_image_store = true;
+
+ info->writes_memory = true;
+ break;
+ case nir_intrinsic_image_deref_store:
+ info->writes_memory = true;
+ break;
+ case nir_intrinsic_bindless_image_atomic_add:
+ case nir_intrinsic_bindless_image_atomic_min:
+ case nir_intrinsic_bindless_image_atomic_max:
+ case nir_intrinsic_bindless_image_atomic_and:
+ case nir_intrinsic_bindless_image_atomic_or:
+ case nir_intrinsic_bindless_image_atomic_xor:
+ case nir_intrinsic_bindless_image_atomic_exchange:
+ case nir_intrinsic_bindless_image_atomic_comp_swap:
+ info->uses_bindless_images = true;
+
+ if (nir_intrinsic_image_dim(intr) == GLSL_SAMPLER_DIM_BUF)
+ info->uses_bindless_buffer_atomic = true;
+ else
+ info->uses_bindless_image_atomic = true;
- if (glsl_get_sampler_dim(image_deref->type) == GLSL_SAMPLER_DIM_BUF)
- info->uses_bindless_buffer_store = true;
- else
- info->uses_bindless_image_store = true;
- }
info->writes_memory = true;
break;
- }
case nir_intrinsic_image_deref_atomic_add:
case nir_intrinsic_image_deref_atomic_min:
case nir_intrinsic_image_deref_atomic_max:
case nir_intrinsic_image_deref_atomic_or:
case nir_intrinsic_image_deref_atomic_xor:
case nir_intrinsic_image_deref_atomic_exchange:
- case nir_intrinsic_image_deref_atomic_comp_swap: {
- nir_variable *var = intrinsic_get_var(intr);
- if (var->data.bindless) {
- info->uses_bindless_images = true;
-
- if (glsl_get_sampler_dim(var->type) == GLSL_SAMPLER_DIM_BUF)
- info->uses_bindless_buffer_atomic = true;
- else
- info->uses_bindless_image_atomic = true;
- }
+ case nir_intrinsic_image_deref_atomic_comp_swap:
info->writes_memory = true;
break;
- }
case nir_intrinsic_store_ssbo:
case nir_intrinsic_ssbo_atomic_add:
case nir_intrinsic_ssbo_atomic_imin:
info->properties[TGSI_PROPERTY_NEXT_SHADER] =
pipe_shader_type_from_mesa(nir->info.next_stage);
+ if (nir->info.stage == MESA_SHADER_VERTEX) {
+ info->properties[TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION] =
+ nir->info.vs.window_space_position;
+ }
+
if (nir->info.stage == MESA_SHADER_TESS_CTRL) {
info->properties[TGSI_PROPERTY_TCS_VERTICES_OUT] =
nir->info.tess.tcs_vertices_out;
/* Fragment shader position is a system value. */
if (nir->info.stage == MESA_SHADER_FRAGMENT &&
variable->data.location == VARYING_SLOT_POS) {
- if (variable->data.pixel_center_integer)
+ if (nir->info.fs.pixel_center_integer)
info->properties[TGSI_PROPERTY_FS_COORD_PIXEL_CENTER] =
TGSI_FS_COORD_PIXEL_CENTER_INTEGER;
struct set *ubo_set = _mesa_set_create(NULL, _mesa_hash_pointer,
_mesa_key_pointer_equal);
+ struct set *ssbo_set = _mesa_set_create(NULL, _mesa_hash_pointer,
+ _mesa_key_pointer_equal);
/* Intialise const_file_max[0] */
info->const_file_max[0] = -1;
+ /* The first 8 are reserved for atomic counters using ssbo */
+ unsigned ssbo_idx = 8;
+
unsigned ubo_idx = 1;
nir_foreach_variable(variable, &nir->uniforms) {
const struct glsl_type *type = variable->type;
enum glsl_base_type base_type =
glsl_get_base_type(glsl_without_array(type));
unsigned aoa_size = MAX2(1, glsl_get_aoa_size(type));
- unsigned loc = variable->data.location;
+ unsigned loc = variable->data.driver_location / 4;
int slot_count = glsl_count_attribute_slots(type, false);
int max_slot = MAX2(info->const_file_max[0], (int) loc) + slot_count;
*/
if (variable->interface_type != NULL) {
if (variable->data.mode == nir_var_uniform ||
- variable->data.mode == nir_var_mem_ubo) {
+ variable->data.mode == nir_var_mem_ubo ||
+ variable->data.mode == nir_var_mem_ssbo) {
+
+ struct set *buf_set = variable->data.mode == nir_var_mem_ssbo ?
+ ssbo_set : ubo_set;
unsigned block_count;
if (base_type != GLSL_TYPE_INTERFACE) {
struct set_entry *entry =
- _mesa_set_search(ubo_set, variable->interface_type);
+ _mesa_set_search(buf_set, variable->interface_type);
/* Check if we have already processed
* a member from this ubo.
block_count = aoa_size;
}
- info->const_buffers_declared |= u_bit_consecutive(ubo_idx, block_count);
- ubo_idx += block_count;
+ if (variable->data.mode == nir_var_uniform ||
+ variable->data.mode == nir_var_mem_ubo) {
+ info->const_buffers_declared |= u_bit_consecutive(ubo_idx, block_count);
+ ubo_idx += block_count;
+ } else {
+ assert(variable->data.mode == nir_var_mem_ssbo);
- _mesa_set_add(ubo_set, variable->interface_type);
- }
+ info->shader_buffers_declared |= u_bit_consecutive(ssbo_idx, block_count);
+ ssbo_idx += block_count;
+ }
- if (variable->data.mode == nir_var_mem_ssbo) {
- /* TODO: make this more accurate */
- info->shader_buffers_declared =
- u_bit_consecutive(0, SI_NUM_SHADER_BUFFERS);
+ _mesa_set_add(buf_set, variable->interface_type);
}
continue;
}
_mesa_set_destroy(ubo_set, NULL);
+ _mesa_set_destroy(ssbo_set, NULL);
info->num_written_clipdistance = nir->info.clip_distance_array_size;
info->num_written_culldistance = nir->info.cull_distance_array_size;
}
}
+void
+si_nir_opts(struct nir_shader *nir)
+{
+ bool progress;
+ unsigned lower_flrp =
+ (nir->options->lower_flrp16 ? 16 : 0) |
+ (nir->options->lower_flrp32 ? 32 : 0) |
+ (nir->options->lower_flrp64 ? 64 : 0);
+
+ do {
+ progress = false;
+
+ NIR_PASS_V(nir, nir_lower_vars_to_ssa);
+
+ NIR_PASS(progress, nir, nir_opt_copy_prop_vars);
+ NIR_PASS(progress, nir, nir_opt_dead_write_vars);
+
+ NIR_PASS_V(nir, nir_lower_alu_to_scalar, NULL);
+ NIR_PASS_V(nir, nir_lower_phis_to_scalar);
+
+ /* (Constant) copy propagation is needed for txf with offsets. */
+ NIR_PASS(progress, nir, nir_copy_prop);
+ NIR_PASS(progress, nir, nir_opt_remove_phis);
+ NIR_PASS(progress, nir, nir_opt_dce);
+ if (nir_opt_trivial_continues(nir)) {
+ progress = true;
+ NIR_PASS(progress, nir, nir_copy_prop);
+ NIR_PASS(progress, nir, nir_opt_dce);
+ }
+ NIR_PASS(progress, nir, nir_opt_if, true);
+ NIR_PASS(progress, nir, nir_opt_dead_cf);
+ NIR_PASS(progress, nir, nir_opt_cse);
+ NIR_PASS(progress, nir, nir_opt_peephole_select, 8, true, true);
+
+ /* Needed for algebraic lowering */
+ NIR_PASS(progress, nir, nir_opt_algebraic);
+ NIR_PASS(progress, nir, nir_opt_constant_folding);
+
+ if (lower_flrp != 0) {
+ bool lower_flrp_progress = false;
+
+ NIR_PASS(lower_flrp_progress, nir, nir_lower_flrp,
+ lower_flrp,
+ false /* always_precise */,
+ nir->options->lower_ffma);
+ if (lower_flrp_progress) {
+ NIR_PASS(progress, nir,
+ nir_opt_constant_folding);
+ progress = true;
+ }
+
+ /* Nothing should rematerialize any flrps, so we only
+ * need to do this lowering once.
+ */
+ lower_flrp = 0;
+ }
+
+ NIR_PASS(progress, nir, nir_opt_undef);
+ NIR_PASS(progress, nir, nir_opt_conditional_discard);
+ if (nir->options->max_unroll_iterations) {
+ NIR_PASS(progress, nir, nir_opt_loop_unroll, 0);
+ }
+ } while (progress);
+}
+
/**
* Perform "lowering" operations on the NIR that are run once when the shader
* selector is created.
* - ensure constant offsets for texture instructions are folded
* and copy-propagated
*/
- NIR_PASS_V(sel->nir, nir_lower_returns);
- NIR_PASS_V(sel->nir, nir_lower_vars_to_ssa);
- NIR_PASS_V(sel->nir, nir_lower_alu_to_scalar);
- NIR_PASS_V(sel->nir, nir_lower_phis_to_scalar);
static const struct nir_lower_tex_options lower_tex_options = {
.lower_txp = ~0u,
ac_lower_indirect_derefs(sel->nir, sel->screen->info.chip_class);
- bool progress;
- do {
- progress = false;
-
- /* (Constant) copy propagation is needed for txf with offsets. */
- NIR_PASS(progress, sel->nir, nir_copy_prop);
- NIR_PASS(progress, sel->nir, nir_opt_remove_phis);
- NIR_PASS(progress, sel->nir, nir_opt_dce);
- if (nir_opt_trivial_continues(sel->nir)) {
- progress = true;
- NIR_PASS(progress, sel->nir, nir_copy_prop);
- NIR_PASS(progress, sel->nir, nir_opt_dce);
- }
- NIR_PASS(progress, sel->nir, nir_opt_if);
- NIR_PASS(progress, sel->nir, nir_opt_dead_cf);
- NIR_PASS(progress, sel->nir, nir_opt_cse);
- NIR_PASS(progress, sel->nir, nir_opt_peephole_select, 8, true, true);
-
- /* Needed for algebraic lowering */
- NIR_PASS(progress, sel->nir, nir_opt_algebraic);
- NIR_PASS(progress, sel->nir, nir_opt_constant_folding);
-
- NIR_PASS(progress, sel->nir, nir_opt_undef);
- NIR_PASS(progress, sel->nir, nir_opt_conditional_discard);
- if (sel->nir->options->max_unroll_iterations) {
- NIR_PASS(progress, sel->nir, nir_opt_loop_unroll, 0);
- }
- } while (progress);
+ si_nir_opts(sel->nir);
NIR_PASS_V(sel->nir, nir_lower_bool_to_int32);
+
+ /* Strip the resulting shader so that the shader cache is more likely
+ * to hit from other similar shaders.
+ */
+ nir_strip(sel->nir);
}
static void declare_nir_input_vs(struct si_shader_context *ctx,