[Mesa-dev] [RFC] main/copyimage: Add check for incomplete src or dst textures

[Eduardo Lima Mitev]

>From OpenGL 4.5 Core Profile spec, section 18.3.2 (Copying Between Images),
page 517 of the PDF:

   "An INVALID_OPERATION error is generated if either object is a texture
    and the texture is not complete (as defined in section 8.17), if the
    source and destination internal formats are not compatible (see below),
    or if the number of samples do not match."

The check for texture completeness is currently missing.

Fixes CTS test 'GL45-CTS.copy_image.incomplete_tex'.

However, this fix brings up a number of regressions for other CTS and
Piglit tests related with glCopyImageSubData, because textures that were
previously assumed complete by a test, don't pass the check now.

For example, Piglit test
'piglit.object namespace pollution.program with glcopyimagesubdata'
$ bin/object-namespace-pollution glCopyImageSubData program -auto -fbo

Analyzing all regressions, my conclusion is that this fix seems legit,
and the tests' textures are incomplete for different reasons:

For example, the Piglit test above simply creates two textures and perform
a glCopyImageSubData between them. But by default these textures are
incomplete because:

* The default minifying filter is NEAREST_MIPMAP_LINEAR (per OpenGL 4.5
spec, section 8.22. 'TEXTURE STATE AND PROXY STATE', page 269 of the PDF).

* Mesa set the default texture MaxLevel to 1000.

The combination of these two values renders the texture mipmap-incomplete.
But if I modify the test and add:

glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);

or

glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, 0);

Then the texture becomes mipmap-complete, passes the check and the test
passes again.

My confusion comes from the source and destination level parameters of
glCopyImageSubData. When these levels are zero, do we still consider the
texture incomplete if it is not mipmap-complete? The spec suggests that
the levels don't affect that check.

So, my questions are:

* Does this patch looks legit? We are definitely not checking texture
  completeness in glCopyImageSubData right now, so at least a variation
  of this patch is necessary.

* Do the level parameters of glCopyImageSubData, assuming they are in
  range, affect the check for mipmap-completeness of a texture?

cheers,

Eduardo
---
 src/mesa/main/copyimage.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/src/mesa/main/copyimage.c b/src/mesa/main/copyimage.c
index cf25159..9fb5ca9 100644
--- a/src/mesa/main/copyimage.c
+++ b/src/mesa/main/copyimage.c
@@ -580,6 +580,25 @@ _mesa_CopyImageSubData(GLuint srcName, GLenum srcTarget, GLint srcLevel,
       return;
    }
 
+   if (srcTarget != GL_RENDERBUFFER) {
+      assert(srcTexImage);
+      if (!_mesa_is_texture_complete(srcTexImage->TexObject,
+                                     &srcTexImage->TexObject->Sampler)) {
+         _mesa_error(ctx, GL_INVALID_OPERATION,
+                     "glCopyImageSubData(incomplete src texture)");
+      }
+   }
+
+   if (dstTarget != GL_RENDERBUFFER) {
+      assert(dstTexImage);
+      if (!_mesa_is_texture_complete(dstTexImage->TexObject,
+                                     &dstTexImage->TexObject->Sampler)) {
+         _mesa_error(ctx, GL_INVALID_OPERATION,
+                     "glCopyImageSubData(incomplete dst texture)");
+         return;
+      }
+   }
+
    /* loop over 2D slices/faces/layers */
    for (i = 0; i < srcDepth; ++i) {
       int newSrcZ = srcZ + i;
-- 
2.7.4