GPU: Add workaround for faulty default attrib values on some drivers

On some drivers, the default values is not respected correctly.

To workaround this we create a small VBO that contains only 1 vec4 worth of
data and just bind it using glBindVertexBuffer to ensure 0 stride.

This fixes T75069 Instances not rendered correctly by workbench.

source/blender/gpu/GPU_shader_interface.h

@@ -88,6 +88,8 @@ typedef struct GPUShaderInterface {
   char *name_buffer;
   struct GPUBatch **batches; /* references to batches using this interface */
   uint batches_len;
+  /** All enabled attribs in this shader. Used to set default values for unbound attribs. */
+  uint16_t enabled_attrib_mask;
 } GPUShaderInterface;
 
 GPUShaderInterface *GPU_shaderinterface_create(int32_t program_id);

source/blender/gpu/intern/gpu_batch.c

@@ -42,6 +42,8 @@
 #include <stdlib.h>
 #include <string.h>
 
+static GLuint g_default_attrib_vbo = 0;
+
 static void batch_update_program_bindings(GPUBatch *batch, uint i_first);
 
 void GPU_batch_vao_cache_clear(GPUBatch *batch)
@@ -414,6 +416,7 @@ void gpu_batch_remove_interface_ref(GPUBatch *batch, const GPUShaderInterface *i
 
 static void create_bindings(GPUVertBuf *verts,
                             const GPUShaderInterface *interface,
+                            uint16_t *attrib_mask,
                             uint v_first,
                             const bool use_instancing)
 {
@@ -446,6 +449,8 @@ static void create_bindings(GPUVertBuf *verts,
         continue;
       }
 
+      *attrib_mask &= ~(1 << input->location);
+
       if (a->comp_len == 16 || a->comp_len == 12 || a->comp_len == 8) {
 #if TRUST_NO_ONE
         assert(a->fetch_mode == GPU_FETCH_FLOAT);
@@ -487,17 +492,35 @@ static void create_bindings(GPUVertBuf *verts,
 
 static void batch_update_program_bindings(GPUBatch *batch, uint i_first)
 {
+  uint16_t attrib_mask = batch->interface->enabled_attrib_mask;
+
   /* Reverse order so first vbos have more prevalence (in term of attrib override). */
   for (int v = GPU_BATCH_VBO_MAX_LEN - 1; v > -1; v--) {
     if (batch->verts[v] != NULL) {
-      create_bindings(batch->verts[v], batch->interface, 0, false);
+      create_bindings(batch->verts[v], batch->interface, &attrib_mask, 0, false);
     }
   }
+
   for (int v = GPU_BATCH_INST_VBO_MAX_LEN - 1; v > -1; v--) {
     if (batch->inst[v]) {
-      create_bindings(batch->inst[v], batch->interface, i_first, true);
+      create_bindings(batch->inst[v], batch->interface, &attrib_mask, i_first, true);
     }
   }
+
+  if (attrib_mask != 0 && GLEW_ARB_vertex_attrib_binding) {
+    for (uint16_t mask = 1, a = 0; a < 16; a++, mask <<= 1) {
+      if (attrib_mask & mask) {
+        /* This replaces glVertexAttrib4f(a, 0.0f, 0.0f, 0.0f, 1.0f); with a more modern style.
+         * Fix issues for some drivers (see T75069). */
+        glBindVertexBuffer(a, g_default_attrib_vbo, (intptr_t)0, (intptr_t)0);
+
+        glEnableVertexAttribArray(a);
+        glVertexAttribFormat(a, 4, GL_FLOAT, GL_FALSE, 0);
+        glVertexAttribBinding(a, a);
+      }
+    }
+  }
+
   if (batch->elem) {
     GPU_indexbuf_use(batch->elem);
   }
@@ -1002,11 +1025,23 @@ void GPU_batch_program_set_imm_shader(GPUBatch *batch)
 
 void gpu_batch_init(void)
 {
+  if (g_default_attrib_vbo == 0) {
+    g_default_attrib_vbo = GPU_buf_alloc();
+
+    float default_attrib_data[4] = {0.0f, 0.0f, 0.0f, 1.0f};
+    glBindBuffer(GL_ARRAY_BUFFER, g_default_attrib_vbo);
+    glBufferData(GL_ARRAY_BUFFER, sizeof(float) * 4, default_attrib_data, GL_STATIC_DRAW);
+    glBindBuffer(GL_ARRAY_BUFFER, 0);
+  }
+
   gpu_batch_presets_init();
 }
 
 void gpu_batch_exit(void)
 {
+  GPU_buf_free(g_default_attrib_vbo);
+  g_default_attrib_vbo = 0;
+
   gpu_batch_presets_exit();
 }
 

source/blender/gpu/intern/gpu_shader_interface.c

@@ -236,6 +236,7 @@ GPUShaderInterface *GPU_shaderinterface_create(int32_t program)
   shaderface->name_buffer = MEM_mallocN(name_buffer_len, "name_buffer");
 
   /* Attributes */
+  shaderface->enabled_attrib_mask = 0;
   for (uint32_t i = 0; i < attr_len; i++) {
     GPUShaderInput *input = MEM_mallocN(sizeof(GPUShaderInput), "GPUShaderInput Attr");
     GLsizei remaining_buffer = name_buffer_len - shaderface->name_buffer_offset;
@@ -255,6 +256,8 @@ GPUShaderInterface *GPU_shaderinterface_create(int32_t program)
 
     input->location = glGetAttribLocation(program, name);
 
+    shaderface->enabled_attrib_mask |= (1 << input->location);
+
     set_input_name(shaderface, input, name, name_len);
 
     shader_input_to_bucket(input, shaderface->attr_buckets);