glsl: fix slow linking of uniforms in the nir linker

Currently the nir linker resizes the amount of storage needed to hold
uniform information on the fly while linking. As shaders can contain
thousands of uniforms this can be very slow. For example some Godot
shaders can take 30 seconds to compile on some machines.

In this change we count the amount of storage needed before we start
processing the uniforms. This is what the GLSL IR linker does also.

Fixes: 95f555a93a88 ("st/glsl_to_nir: make use of nir linker for linking uniforms")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2996
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5137>

diff --git a/src/compiler/glsl/gl_nir_link_uniforms.c b/src/compiler/glsl/gl_nir_link_uniforms.c
index ee6e2395eea1e19b0b070554e03884b0a9719c5b..c8d32bdf3384668ef109c0afdfb54328ccb9db94 100644 (file)
--- a/src/compiler/glsl/gl_nir_link_uniforms.c
+++ b/src/compiler/glsl/gl_nir_link_uniforms.c
@@ -45,6 +45,34 @@ is_gl_identifier(const char *s)
    return s && s[0] == 'g' && s[1] == 'l' && s[2] == '_';
 }
 
+static unsigned
+uniform_storage_size(const struct glsl_type *type)
+{
+   switch (glsl_get_base_type(type)) {
+   case GLSL_TYPE_STRUCT:
+   case GLSL_TYPE_INTERFACE: {
+      unsigned size = 0;
+      for (unsigned i = 0; i < glsl_get_length(type); i++)
+         size += uniform_storage_size(glsl_get_struct_field(type, i));
+      return size;
+   }
+   case GLSL_TYPE_ARRAY: {
+      const struct glsl_type *e_type = glsl_get_array_element(type);
+      enum glsl_base_type e_base_type = glsl_get_base_type(e_type);
+      if (e_base_type == GLSL_TYPE_STRUCT ||
+          e_base_type == GLSL_TYPE_INTERFACE ||
+          e_base_type == GLSL_TYPE_ARRAY) {
+         unsigned length = !glsl_type_is_unsized_array(type) ?
+            glsl_get_length(type) : 1;
+         return length * uniform_storage_size(e_type);
+      } else
+         return 1;
+   }
+   default:
+      return 1;
+   }
+}
+
 static void
 nir_setup_uniform_remap_tables(struct gl_context *ctx,
                                struct gl_shader_program *prog)
@@ -1142,15 +1170,20 @@ nir_link_uniform(struct gl_context *ctx,
 
       return location_count;
    } else {
-      /* Create a new uniform storage entry */
-      prog->data->UniformStorage =
-         reralloc(prog->data,
-                  prog->data->UniformStorage,
-                  struct gl_uniform_storage,
-                  prog->data->NumUniformStorage + 1);
-      if (!prog->data->UniformStorage) {
-         linker_error(prog, "Out of memory during linking.\n");
-         return -1;
+      /* TODO: reallocating storage is slow, we should figure out a way to
+       * allocate storage up front for spirv like we do for GLSL.
+       */
+      if (prog->data->spirv) {
+         /* Create a new uniform storage entry */
+         prog->data->UniformStorage =
+            reralloc(prog->data,
+                     prog->data->UniformStorage,
+                     struct gl_uniform_storage,
+                     prog->data->NumUniformStorage + 1);
+         if (!prog->data->UniformStorage) {
+            linker_error(prog, "Out of memory during linking.\n");
+            return -1;
+         }
       }
 
       uniform = &prog->data->UniformStorage[prog->data->NumUniformStorage];
@@ -1349,6 +1382,43 @@ gl_nir_link_uniforms(struct gl_context *ctx,
    prog->data->UniformStorage = NULL;
    prog->data->NumUniformStorage = 0;
 
+   /* Count total number of uniforms and allocate storage */
+   unsigned storage_size = 0;
+   if (!prog->data->spirv) {
+      struct set *storage_counted =
+         _mesa_set_create(NULL, _mesa_hash_string, _mesa_key_string_equal);
+      for (unsigned stage = 0; stage < MESA_SHADER_STAGES; stage++) {
+         struct gl_linked_shader *sh = prog->_LinkedShaders[stage];
+         if (!sh)
+            continue;
+
+         nir_foreach_variable(var, &sh->Program->nir->uniforms) {
+            const struct glsl_type *type = var->type;
+            const char *name = var->name;
+            if (nir_variable_is_in_block(var) &&
+                glsl_without_array(type) == var->interface_type) {
+               type = glsl_without_array(var->type);
+               name = glsl_get_type_name(type);
+            }
+
+            struct set_entry *entry = _mesa_set_search(storage_counted, name);
+            if (!entry) {
+               storage_size += uniform_storage_size(type);
+               _mesa_set_add(storage_counted, name);
+            }
+         }
+      }
+      _mesa_set_destroy(storage_counted, NULL);
+
+      prog->data->UniformStorage = rzalloc_array(prog->data,
+                                                 struct gl_uniform_storage,
+                                                 storage_size);
+      if (!prog->data->UniformStorage) {
+         linker_error(prog, "Out of memory while linking uniforms.\n");
+         return false;
+      }
+   }
+
    /* Iterate through all linked shaders */
    struct nir_link_uniforms_state state = {0,};
    state.uniform_hash = _mesa_hash_table_create(NULL, _mesa_hash_string,
@@ -1639,6 +1709,8 @@ gl_nir_link_uniforms(struct gl_context *ctx,
    prog->data->NumHiddenUniforms = state.num_hidden_uniforms;
    prog->data->NumUniformDataSlots = state.num_values;
 
+   assert(prog->data->spirv || prog->data->NumUniformStorage == storage_size);
+
    if (prog->data->spirv)
       prog->NumUniformRemapTable = state.max_uniform_location;