DRW-Next: Add uniform attributes (object attributes) support

This replaces the direct shader uniform layout declaration by a linear
search through a global buffer.

Each instance has an attribute offset inside the global buffer and an
attribute count.

This removes any padding and tighly pack all uniform attributes inside
a single buffer.

This would also remove the limit of 8 attribute but it is kept because of
compatibility with the old system that is still used by the old draw
manager.

4 files changed, 29 insertions, 21 deletions

diff --git a/source/blender/gpu/GPU_material.h b/source/blender/gpu/GPU_material.h
index 042979b3a86..023221543ec 100644
--- a/source/blender/gpu/GPU_material.h
+++ b/source/blender/gpu/GPU_material.h
@@ -149,7 +149,10 @@ GPUNodeLink *GPU_attribute_with_default(GPUMaterial *mat,
                                         eCustomDataType type,
                                         const char *name,
                                         eGPUDefaultValue default_value);
-GPUNodeLink *GPU_uniform_attribute(GPUMaterial *mat, const char *name, bool use_dupli);
+GPUNodeLink *GPU_uniform_attribute(GPUMaterial *mat,
+                                   const char *name,
+                                   bool use_dupli,
+                                   uint32_t *r_hash);
 GPUNodeLink *GPU_image(GPUMaterial *mat,
                        struct Image *ima,
                        struct ImageUser *iuser,
diff --git a/source/blender/gpu/intern/gpu_codegen.cc b/source/blender/gpu/intern/gpu_codegen.cc
index d58ede4ccd8..f774f33e03d 100644
--- a/source/blender/gpu/intern/gpu_codegen.cc
+++ b/source/blender/gpu/intern/gpu_codegen.cc
@@ -199,8 +199,7 @@ static std::ostream &operator<<(std::ostream &stream, const GPUOutput *output)
 }
 
 /* Trick type to change overload and keep a somewhat nice syntax. */
-struct GPUConstant : public GPUInput {
-};
+struct GPUConstant : public GPUInput {};
 
 /* Print data constructor (i.e: vec2(1.0f, 1.0f)). */
 static std::ostream &operator<<(std::ostream &stream, const GPUConstant *input)
@@ -208,9 +207,10 @@ static std::ostream &operator<<(std::ostream &stream, const GPUConstant *input)
   stream << input->type << "(";
   for (int i = 0; i < input->type; i++) {
     char formated_float[32];
-    /* Print with the maximum precision for single precision float using scientific notation.
-     * See https://stackoverflow.com/questions/16839658/#answer-21162120 */
-    SNPRINTF(formated_float, "%.9g", input->vec[i]);
+    /* Use uint representation to allow exact same bit pattern even if NaN. This is because we can
+     * pass UINTs as floats for constants. */
+    const uint32_t *uint_vec = reinterpret_cast<const uint32_t *>(input->vec);
+    SNPRINTF(formated_float, "uintBitsToFloat(%uu)", uint_vec[i]);
     stream << formated_float;
     if (i < input->type - 1) {
       stream << ", ";
diff --git a/source/blender/gpu/intern/gpu_node_graph.c b/source/blender/gpu/intern/gpu_node_graph.c
index 4d391ff9063..f82af7538b5 100644
--- a/source/blender/gpu/intern/gpu_node_graph.c
+++ b/source/blender/gpu/intern/gpu_node_graph.c
@@ -320,20 +320,7 @@ void gpu_node_graph_finalize_uniform_attrs(GPUNodeGraph *graph)
 
   LISTBASE_FOREACH (GPUUniformAttr *, attr, &attrs->list) {
     attr->id = next_id++;
-
-    attr->hash_code = BLI_ghashutil_strhash_p(attr->name);
-
-    if (attr->use_dupli) {
-      attr->hash_code ^= BLI_ghashutil_uinthash(attr->id);
-    }
-
-    attrs->hash_code ^= attr->hash_code;
-
-    {
-      char attr_name_esc[sizeof(attr->name) * 2];
-      BLI_str_escape(attr_name_esc, attr->name, sizeof(attr_name_esc));
-      SNPRINTF(attr->name_id_prop, "[\"%s\"]", attr_name_esc);
-    }
+    attrs->hash_code ^= BLI_ghashutil_uinthash(attr->hash_code + (1 << (attr->id + 1)));
   }
 }
 
@@ -428,7 +415,13 @@ static GPUUniformAttr *gpu_node_graph_add_uniform_attribute(GPUNodeGraph *graph,
   if (attr == NULL && attrs->count < GPU_MAX_UNIFORM_ATTR) {
     attr = MEM_callocN(sizeof(*attr), __func__);
     STRNCPY(attr->name, name);
+    {
+      char attr_name_esc[sizeof(attr->name) * 2];
+      BLI_str_escape(attr_name_esc, attr->name, sizeof(attr_name_esc));
+      SNPRINTF(attr->name_id_prop, "[\"%s\"]", attr_name_esc);
+    }
     attr->use_dupli = use_dupli;
+    attr->hash_code = BLI_ghashutil_strhash_p(attr->name) << 1 | (attr->use_dupli ? 0 : 1);
     attr->id = -1;
     BLI_addtail(&attrs->list, attr);
     attrs->count++;
@@ -532,16 +525,21 @@ GPUNodeLink *GPU_attribute_with_default(GPUMaterial *mat,
   return link;
 }
 
-GPUNodeLink *GPU_uniform_attribute(GPUMaterial *mat, const char *name, bool use_dupli)
+GPUNodeLink *GPU_uniform_attribute(GPUMaterial *mat,
+                                   const char *name,
+                                   bool use_dupli,
+                                   uint32_t *r_hash)
 {
   GPUNodeGraph *graph = gpu_material_node_graph(mat);
   GPUUniformAttr *attr = gpu_node_graph_add_uniform_attribute(graph, name, use_dupli);
 
   /* Dummy fallback if out of slots. */
   if (attr == NULL) {
+    *r_hash = 0;
     static const float zero_data[GPU_MAX_CONSTANT_DATA] = {0.0f};
     return GPU_constant(zero_data);
   }
+  *r_hash = attr->hash_code;
 
   GPUNodeLink *link = gpu_node_link_create();
   link->link_type = GPU_NODE_LINK_UNIFORM_ATTR;
diff --git a/source/blender/gpu/shaders/material/gpu_shader_material_attribute.glsl b/source/blender/gpu/shaders/material/gpu_shader_material_attribute.glsl
index af4a511d627..bacf089deb1 100644
--- a/source/blender/gpu/shaders/material/gpu_shader_material_attribute.glsl
+++ b/source/blender/gpu/shaders/material/gpu_shader_material_attribute.glsl
@@ -22,6 +22,13 @@ void node_attribute_flame(vec4 attr, out float out_attr)
   out_attr = attr.x;
 }
 
+void node_attribute_uniform(vec4 attr, const float attr_hash, out vec4 out_attr)
+{
+  /* Temporary solution to support both old UBO attribs and new SSBO loading.
+   * Old UBO load is already done through `attr` and will just be passed through. */
+  out_attr = attr_load_uniform(attr, floatBitsToUint(attr_hash));
+}
+
 void node_attribute(
     vec4 attr, out vec4 outcol, out vec3 outvec, out float outf, out float outalpha)
 {