type = glsl_get_array_element(type);
}
+ if (var->data.per_view) {
+ /* TODO: Per view and Per Vertex are not currently used together. When
+ * they start to be used (e.g. when adding Primitive Replication for GS
+ * on Intel), verify that "peeling" the type twice is correct. This
+ * assert ensures we remember it.
+ */
+ assert(!nir_is_per_vertex_io(var, shader->info.stage));
+ assert(glsl_type_is_array(type));
+ type = glsl_get_array_element(type);
+ }
+
const unsigned slots =
var->data.compact ? DIV_ROUND_UP(glsl_get_length(type), 4)
: glsl_count_attribute_slots(type, false);
type = glsl_get_array_element(type);
}
+ /* Per view variables will be considered as a whole. */
+ if (var->data.per_view)
+ return false;
+
/* The code below only handles:
*
* - Indexing into matrices
return true;
}
+static void
+update_memory_written_for_deref(nir_shader *shader, nir_deref_instr *deref)
+{
+ switch (deref->mode) {
+ case nir_var_mem_ssbo:
+ case nir_var_mem_global:
+ shader->info.writes_memory = true;
+ break;
+ default:
+ /* Nothing to do. */
+ break;
+ }
+}
+
static void
gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader,
void *dead_ctx)
{
+ unsigned slot_mask = 0;
+
+ if (nir_intrinsic_infos[instr->intrinsic].index_map[NIR_INTRINSIC_IO_SEMANTICS] > 0) {
+ nir_io_semantics semantics = nir_intrinsic_io_semantics(instr);
+
+ slot_mask = BITFIELD64_RANGE(semantics.location, semantics.num_slots);
+ }
+
switch (instr->intrinsic) {
case nir_intrinsic_demote:
case nir_intrinsic_demote_if:
shader->info.fs.uses_demote = true;
- /* fallthrough: quads with helper lanes only might be discarded entirely */
+ /* fallthrough - quads with helper lanes only might be discarded entirely */
case nir_intrinsic_discard:
case nir_intrinsic_discard_if:
- assert(shader->info.stage == MESA_SHADER_FRAGMENT);
- shader->info.fs.uses_discard = true;
+ /* Freedreno uses the discard_if intrinsic to end GS invocations that
+ * don't produce a vertex, so we only set uses_discard if executing on
+ * a fragment shader. */
+ if (shader->info.stage == MESA_SHADER_FRAGMENT)
+ shader->info.fs.uses_discard = true;
break;
case nir_intrinsic_interp_deref_at_centroid:
}
}
}
+ if (instr->intrinsic == nir_intrinsic_store_deref)
+ update_memory_written_for_deref(shader, deref);
break;
}
+ case nir_intrinsic_load_input:
+ if (shader->info.stage == MESA_SHADER_TESS_EVAL)
+ shader->info.patch_inputs_read |= slot_mask;
+ else
+ shader->info.inputs_read |= slot_mask;
+ break;
+
+ case nir_intrinsic_load_per_vertex_input:
+ case nir_intrinsic_load_input_vertex:
+ case nir_intrinsic_load_interpolated_input:
+ shader->info.inputs_read |= slot_mask;
+ break;
+
+ case nir_intrinsic_load_output:
+ if (shader->info.stage == MESA_SHADER_TESS_CTRL)
+ shader->info.patch_outputs_read |= slot_mask;
+ else
+ shader->info.outputs_read |= slot_mask;
+ break;
+
+ case nir_intrinsic_load_per_vertex_output:
+ shader->info.outputs_read |= slot_mask;
+ break;
+
+ case nir_intrinsic_store_output:
+ if (shader->info.stage == MESA_SHADER_TESS_CTRL)
+ shader->info.patch_outputs_written |= slot_mask;
+ else
+ shader->info.outputs_written |= slot_mask;
+ break;
+
+ case nir_intrinsic_store_per_vertex_output:
+ shader->info.outputs_written |= slot_mask;
+ break;
+
case nir_intrinsic_load_draw_id:
case nir_intrinsic_load_frag_coord:
case nir_intrinsic_load_point_coord:
+ case nir_intrinsic_load_line_coord:
case nir_intrinsic_load_front_face:
case nir_intrinsic_load_vertex_id:
case nir_intrinsic_load_vertex_id_zero_base:
case nir_intrinsic_emit_vertex:
case nir_intrinsic_emit_vertex_with_counter:
- if (nir_intrinsic_stream_id(instr) > 0)
- shader->info.gs.uses_streams = true;
+ shader->info.gs.active_stream_mask |= 1 << nir_intrinsic_stream_id(instr);
break;
+ case nir_intrinsic_atomic_counter_inc:
+ case nir_intrinsic_atomic_counter_inc_deref:
+ case nir_intrinsic_atomic_counter_add:
+ case nir_intrinsic_atomic_counter_add_deref:
+ case nir_intrinsic_atomic_counter_pre_dec:
+ case nir_intrinsic_atomic_counter_pre_dec_deref:
+ case nir_intrinsic_atomic_counter_post_dec:
+ case nir_intrinsic_atomic_counter_post_dec_deref:
+ case nir_intrinsic_atomic_counter_min:
+ case nir_intrinsic_atomic_counter_min_deref:
+ case nir_intrinsic_atomic_counter_max:
+ case nir_intrinsic_atomic_counter_max_deref:
+ case nir_intrinsic_atomic_counter_and:
+ case nir_intrinsic_atomic_counter_and_deref:
+ case nir_intrinsic_atomic_counter_or:
+ case nir_intrinsic_atomic_counter_or_deref:
+ case nir_intrinsic_atomic_counter_xor:
+ case nir_intrinsic_atomic_counter_xor_deref:
+ case nir_intrinsic_atomic_counter_exchange:
+ case nir_intrinsic_atomic_counter_exchange_deref:
+ case nir_intrinsic_atomic_counter_comp_swap:
+ case nir_intrinsic_atomic_counter_comp_swap_deref:
case nir_intrinsic_bindless_image_atomic_add:
case nir_intrinsic_bindless_image_atomic_and:
case nir_intrinsic_bindless_image_atomic_comp_swap:
shader->info.writes_memory = true;
break;
+ 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:
+ update_memory_written_for_deref(shader, nir_src_as_deref(instr->src[0]));
+ break;
+
default:
break;
}
{
shader->info.num_textures = 0;
shader->info.num_images = 0;
- shader->info.last_msaa_image = -1;
+ shader->info.image_buffers = 0;
+ shader->info.msaa_images = 0;
- nir_foreach_variable(var, &shader->uniforms) {
- /* Bindless textures and images don't use non-bindless slots. */
- if (var->data.bindless)
+ nir_foreach_uniform_variable(var, shader) {
+ /* Bindless textures and images don't use non-bindless slots.
+ * Interface blocks imply inputs, outputs, UBO, or SSBO, which can only
+ * mean bindless.
+ */
+ if (var->data.bindless || var->interface_type)
continue;
shader->info.num_textures += glsl_type_get_sampler_count(var->type);
- shader->info.num_images += glsl_type_get_image_count(var->type);
- /* Assuming image slots don't have holes (e.g. OpenGL) */
- if (glsl_type_is_image(var->type) &&
- glsl_get_sampler_dim(var->type) == GLSL_SAMPLER_DIM_MS)
- shader->info.last_msaa_image = shader->info.num_images - 1;
+ unsigned num_image_slots = glsl_type_get_image_count(var->type);
+ if (num_image_slots) {
+ const struct glsl_type *image_type = glsl_without_array(var->type);
+
+ if (glsl_get_sampler_dim(image_type) == GLSL_SAMPLER_DIM_BUF) {
+ shader->info.image_buffers |=
+ BITFIELD_RANGE(shader->info.num_images, num_image_slots);
+ }
+ if (glsl_get_sampler_dim(image_type) == GLSL_SAMPLER_DIM_MS) {
+ shader->info.msaa_images |=
+ BITFIELD_RANGE(shader->info.num_images, num_image_slots);
+ }
+ shader->info.num_images += num_image_slots;
+ }
}
shader->info.inputs_read = 0;