X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fcompiler%2Fnir%2Fnir_gather_info.c;h=b6e6b0afd1f2da62c23445645302d6db8fef8778;hb=fd73ed1bd7e4d5a6af05d29908d6c4e2cca868b5;hp=4f4b650ab7ed18512eb8eb0e90c1ca8c82e8f50f;hpb=d80c342d898275cbd6266c37e70dc422590d7d8c;p=mesa.git

diff --git a/src/compiler/nir/nir_gather_info.c b/src/compiler/nir/nir_gather_info.c
index 4f4b650ab7e..b6e6b0afd1f 100644
--- a/src/compiler/nir/nir_gather_info.c
+++ b/src/compiler/nir/nir_gather_info.c
@@ -54,11 +54,6 @@ set_io_mask(nir_shader *shader, nir_variable *var, int offset, int len,
          else
             shader->info.inputs_read |= bitfield;
 
-         /* double inputs read is only for vertex inputs */
-         if (shader->info.stage == MESA_SHADER_VERTEX &&
-             glsl_type_is_dual_slot(glsl_without_array(var->type)))
-            shader->info.vs.double_inputs_read |= bitfield;
-
          if (shader->info.stage == MESA_SHADER_FRAGMENT) {
             shader->info.fs.uses_sample_qualifier |= var->data.sample;
          }
@@ -93,43 +88,31 @@ static void
 mark_whole_variable(nir_shader *shader, nir_variable *var, bool is_output_read)
 {
    const struct glsl_type *type = var->type;
-   bool is_vertex_input = false;
 
    if (nir_is_per_vertex_io(var, shader->info.stage)) {
       assert(glsl_type_is_array(type));
       type = glsl_get_array_element(type);
    }
 
-   if (!shader->options->vs_inputs_dual_locations &&
-       shader->info.stage == MESA_SHADER_VERTEX &&
-       var->data.mode == nir_var_shader_in)
-      is_vertex_input = true;
-
    const unsigned slots =
       var->data.compact ? DIV_ROUND_UP(glsl_get_length(type), 4)
-                        : glsl_count_attribute_slots(type, is_vertex_input);
+                        : glsl_count_attribute_slots(type, false);
 
    set_io_mask(shader, var, 0, slots, is_output_read);
 }
 
 static unsigned
-get_io_offset(nir_deref_var *deref, bool is_vertex_input)
+get_io_offset(nir_deref_instr *deref, bool is_vertex_input)
 {
    unsigned offset = 0;
 
-   nir_deref *tail = &deref->deref;
-   while (tail->child != NULL) {
-      tail = tail->child;
-
-      if (tail->deref_type == nir_deref_type_array) {
-         nir_deref_array *deref_array = nir_deref_as_array(tail);
-
-         if (deref_array->deref_array_type == nir_deref_array_type_indirect) {
+   for (nir_deref_instr *d = deref; d; d = nir_deref_instr_parent(d)) {
+      if (d->deref_type == nir_deref_type_array) {
+         if (!nir_src_is_const(d->arr.index))
             return -1;
-         }
 
-         offset += glsl_count_attribute_slots(tail->type, is_vertex_input) *
-            deref_array->base_offset;
+         offset += glsl_count_attribute_slots(d->type, is_vertex_input) *
+                   nir_src_as_uint(d->arr.index);
       }
       /* TODO: we can get the offset for structs here see nir_lower_io() */
    }
@@ -145,9 +128,9 @@ get_io_offset(nir_deref_var *deref, bool is_vertex_input)
  * occurs, then nothing will be marked and false will be returned.
  */
 static bool
-try_mask_partial_io(nir_shader *shader, nir_deref_var *deref, bool is_output_read)
+try_mask_partial_io(nir_shader *shader, nir_variable *var,
+                    nir_deref_instr *deref, bool is_output_read)
 {
-   nir_variable *var = deref->var;
    const struct glsl_type *type = var->type;
 
    if (nir_is_per_vertex_io(var, shader->info.stage)) {
@@ -174,13 +157,7 @@ try_mask_partial_io(nir_shader *shader, nir_deref_var *deref, bool is_output_rea
       return false;
    }
 
-   bool is_vertex_input = false;
-   if (!shader->options->vs_inputs_dual_locations &&
-       shader->info.stage == MESA_SHADER_VERTEX &&
-       var->data.mode == nir_var_shader_in)
-      is_vertex_input = true;
-
-   unsigned offset = get_io_offset(deref, is_vertex_input);
+   unsigned offset = get_io_offset(deref, false);
    if (offset == -1)
       return false;
 
@@ -196,10 +173,8 @@ try_mask_partial_io(nir_shader *shader, nir_deref_var *deref, bool is_output_rea
    }
 
    /* double element width for double types that takes two slots */
-   if (!is_vertex_input &&
-       glsl_type_is_dual_slot(glsl_without_array(type))) {
+   if (glsl_type_is_dual_slot(glsl_without_array(type)))
       elem_width *= 2;
-   }
 
    if (offset >= num_elems * elem_width * mat_cols) {
       /* Constant index outside the bounds of the matrix/array.  This could
@@ -219,30 +194,33 @@ try_mask_partial_io(nir_shader *shader, nir_deref_var *deref, bool is_output_rea
 }
 
 static void
-gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader)
+gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader,
+                      void *dead_ctx)
 {
    switch (instr->intrinsic) {
+   case nir_intrinsic_demote:
+   case nir_intrinsic_demote_if:
    case nir_intrinsic_discard:
    case nir_intrinsic_discard_if:
       assert(shader->info.stage == MESA_SHADER_FRAGMENT);
       shader->info.fs.uses_discard = true;
       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_load_var:
-   case nir_intrinsic_store_var: {
-      nir_variable *var = instr->variables[0]->var;
-
-      if (var->data.mode == nir_var_shader_in ||
-          var->data.mode == nir_var_shader_out) {
+   case nir_intrinsic_interp_deref_at_centroid:
+   case nir_intrinsic_interp_deref_at_sample:
+   case nir_intrinsic_interp_deref_at_offset:
+   case nir_intrinsic_load_deref:
+   case nir_intrinsic_store_deref:{
+      nir_deref_instr *deref = nir_src_as_deref(instr->src[0]);
+      if (deref->mode == nir_var_shader_in ||
+          deref->mode == nir_var_shader_out) {
+         nir_variable *var = nir_deref_instr_get_variable(deref);
          bool is_output_read = false;
          if (var->data.mode == nir_var_shader_out &&
-             instr->intrinsic == nir_intrinsic_load_var)
+             instr->intrinsic == nir_intrinsic_load_deref)
             is_output_read = true;
 
-         if (!try_mask_partial_io(shader, instr->variables[0], is_output_read))
+         if (!try_mask_partial_io(shader, var, deref, is_output_read))
             mark_whole_variable(shader, var, is_output_read);
 
          /* We need to track which input_reads bits correspond to a
@@ -261,6 +239,7 @@ gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader)
 
    case nir_intrinsic_load_draw_id:
    case nir_intrinsic_load_frag_coord:
+   case nir_intrinsic_load_point_coord:
    case nir_intrinsic_load_front_face:
    case nir_intrinsic_load_vertex_id:
    case nir_intrinsic_load_vertex_id_zero_base:
@@ -286,10 +265,24 @@ gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader)
          (1ull << nir_system_value_from_intrinsic(instr->intrinsic));
       break;
 
+   case nir_intrinsic_quad_broadcast:
+   case nir_intrinsic_quad_swap_horizontal:
+   case nir_intrinsic_quad_swap_vertical:
+   case nir_intrinsic_quad_swap_diagonal:
+      if (shader->info.stage == MESA_SHADER_FRAGMENT)
+         shader->info.fs.needs_helper_invocations = true;
+      break;
+
    case nir_intrinsic_end_primitive:
    case nir_intrinsic_end_primitive_with_counter:
       assert(shader->info.stage == MESA_SHADER_GEOMETRY);
       shader->info.gs.uses_end_primitive = 1;
+      /* fall through */
+
+   case nir_intrinsic_emit_vertex:
+      if (nir_intrinsic_stream_id(instr) > 0)
+         shader->info.gs.uses_streams = true;
+
       break;
 
    default:
@@ -300,17 +293,14 @@ gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader)
 static void
 gather_tex_info(nir_tex_instr *instr, nir_shader *shader)
 {
+   if (shader->info.stage == MESA_SHADER_FRAGMENT &&
+       nir_tex_instr_has_implicit_derivative(instr))
+      shader->info.fs.needs_helper_invocations = true;
+
    switch (instr->op) {
    case nir_texop_tg4:
       shader->info.uses_texture_gather = true;
       break;
-   case nir_texop_txf:
-   case nir_texop_txf_ms:
-   case nir_texop_txf_ms_mcs:
-      shader->info.textures_used_by_txf |=
-         ((1 << MAX2(instr->texture_array_size, 1)) - 1) <<
-         instr->texture_index;
-      break;
    default:
       break;
    }
@@ -323,14 +313,27 @@ gather_alu_info(nir_alu_instr *instr, nir_shader *shader)
    case nir_op_fddx:
    case nir_op_fddy:
       shader->info.uses_fddx_fddy = true;
+      /* Fall through */
+   case nir_op_fddx_fine:
+   case nir_op_fddy_fine:
+   case nir_op_fddx_coarse:
+   case nir_op_fddy_coarse:
+      if (shader->info.stage == MESA_SHADER_FRAGMENT)
+         shader->info.fs.needs_helper_invocations = true;
       break;
    default:
       break;
    }
+
+   shader->info.uses_64bit |= instr->dest.dest.ssa.bit_size == 64;
+   unsigned num_srcs = nir_op_infos[instr->op].num_inputs;
+   for (unsigned i = 0; i < num_srcs; i++) {
+      shader->info.uses_64bit |= nir_src_bit_size(instr->src[i].src) == 64;
+   }
 }
 
 static void
-gather_info_block(nir_block *block, nir_shader *shader)
+gather_info_block(nir_block *block, nir_shader *shader, void *dead_ctx)
 {
    nir_foreach_instr(instr, block) {
       switch (instr->type) {
@@ -338,7 +341,7 @@ gather_info_block(nir_block *block, nir_shader *shader)
          gather_alu_info(nir_instr_as_alu(instr), shader);
          break;
       case nir_instr_type_intrinsic:
-         gather_intrinsic_info(nir_instr_as_intrinsic(instr), shader);
+         gather_intrinsic_info(nir_instr_as_intrinsic(instr), shader, dead_ctx);
          break;
       case nir_instr_type_tex:
          gather_tex_info(nir_instr_as_tex(instr), shader);
@@ -352,53 +355,9 @@ gather_info_block(nir_block *block, nir_shader *shader)
    }
 }
 
-static unsigned
-glsl_type_get_sampler_count(const struct glsl_type *type)
-{
-   if (glsl_type_is_array(type)) {
-      return (glsl_get_aoa_size(type) *
-              glsl_type_get_sampler_count(glsl_without_array(type)));
-   }
-
-   if (glsl_type_is_struct(type)) {
-      unsigned count = 0;
-      for (int i = 0; i < glsl_get_length(type); i++)
-         count += glsl_type_get_sampler_count(glsl_get_struct_field(type, i));
-      return count;
-   }
-
-   if (glsl_type_is_sampler(type))
-      return 1;
-
-   return 0;
-}
-
-static unsigned
-glsl_type_get_image_count(const struct glsl_type *type)
-{
-   if (glsl_type_is_array(type)) {
-      return (glsl_get_aoa_size(type) *
-              glsl_type_get_image_count(glsl_without_array(type)));
-   }
-
-   if (glsl_type_is_struct(type)) {
-      unsigned count = 0;
-      for (int i = 0; i < glsl_get_length(type); i++)
-         count += glsl_type_get_image_count(glsl_get_struct_field(type, i));
-      return count;
-   }
-
-   if (glsl_type_is_image(type))
-      return 1;
-
-   return 0;
-}
-
 void
 nir_shader_gather_info(nir_shader *shader, nir_function_impl *entrypoint)
 {
-   nir_assert_lowered_derefs(shader, nir_lower_load_store_derefs | nir_lower_interp_derefs);
-
    shader->info.num_textures = 0;
    shader->info.num_images = 0;
    nir_foreach_variable(var, &shader->uniforms) {
@@ -415,12 +374,14 @@ nir_shader_gather_info(nir_shader *shader, nir_function_impl *entrypoint)
    shader->info.system_values_read = 0;
    if (shader->info.stage == MESA_SHADER_VERTEX) {
       shader->info.vs.double_inputs = 0;
-      shader->info.vs.double_inputs_read = 0;
    }
    if (shader->info.stage == MESA_SHADER_FRAGMENT) {
       shader->info.fs.uses_sample_qualifier = false;
    }
+
+   void *dead_ctx = ralloc_context(NULL);
    nir_foreach_block(block, entrypoint) {
-      gather_info_block(block, shader);
+      gather_info_block(block, shader, dead_ctx);
    }
+   ralloc_free(dead_ctx);
 }