[Mesa-dev] [PATCH v2] mesa: error out in indirect draw when vertex bindings mismatch

[Tapani Pälli]

On 11/23/2015 11:17 AM, Fredrik Höglund wrote:
> On Thursday 19 November 2015, Tapani Pälli wrote:
>> Patch adds additional mask for tracking which vertex buffer bindings
>> are set. This array can be directly compared to which vertex arrays
>> are enabled and should match when drawing.
>>
>> Fixes following CTS tests:
>>
>>     ES31-CTS.draw_indirect.negative-noVBO-arrays
>>     ES31-CTS.draw_indirect.negative-noVBO-elements
>>
>> v2: update mask in vertex_array_attrib_binding
>>
>> Signed-off-by: Tapani Pälli <tapani.palli at intel.com>
>> ---
>>   src/mesa/main/api_validate.c | 13 +++++++++++++
>>   src/mesa/main/mtypes.h       |  3 +++
>>   src/mesa/main/varray.c       |  7 +++++++
>>   3 files changed, 23 insertions(+)
>>
>> diff --git a/src/mesa/main/api_validate.c b/src/mesa/main/api_validate.c
>> index a490189..e82e89a 100644
>> --- a/src/mesa/main/api_validate.c
>> +++ b/src/mesa/main/api_validate.c
>> @@ -710,6 +710,19 @@ valid_draw_indirect(struct gl_context *ctx,
>>         return GL_FALSE;
>>      }
>>
>> +   /* From OpenGL ES 3.1 spec. section 10.5:
>> +    *     "An INVALID_OPERATION error is generated if zero is bound to
>> +    *     VERTEX_ARRAY_BINDING, DRAW_INDIRECT_BUFFER or to any enabled
>> +    *     vertex array."
>> +    *
>> +    * Here we check that vertex buffer bindings match with enabled
>> +    * vertex arrays.
>> +    */
>> +   if (ctx->Array.VAO->_Enabled != ctx->Array.VAO->VertexBindingMask) {
>> +      _mesa_error(ctx, GL_INVALID_OPERATION, "%s(No VBO bound)", name);
>> +      return GL_FALSE;
>> +   }
>> +
>>      if (!_mesa_valid_prim_mode(ctx, mode, name))
>>         return GL_FALSE;
>>
>> diff --git a/src/mesa/main/mtypes.h b/src/mesa/main/mtypes.h
>> index 4efdf1e..6c6187f 100644
>> --- a/src/mesa/main/mtypes.h
>> +++ b/src/mesa/main/mtypes.h
>> @@ -1419,6 +1419,9 @@ struct gl_vertex_array_object
>>      /** Vertex buffer bindings */
>>      struct gl_vertex_buffer_binding VertexBinding[VERT_ATTRIB_MAX];
>>
>> +   /** Mask indicating which binding points are set. */
>> +   GLbitfield64 VertexBindingMask;
>> +
>>      /** Mask of VERT_BIT_* values indicating which arrays are enabled */
>>      GLbitfield64 _Enabled;
>>
>> diff --git a/src/mesa/main/varray.c b/src/mesa/main/varray.c
>> index 887d0c0..e24342a 100644
>> --- a/src/mesa/main/varray.c
>> +++ b/src/mesa/main/varray.c
>> @@ -174,6 +174,11 @@ bind_vertex_buffer(struct gl_context *ctx,
>>         binding->Offset = offset;
>>         binding->Stride = stride;
>>
>> +      if (vbo == ctx->Shared->NullBufferObj)
>> +         vao->VertexBindingMask &= ~VERT_BIT(index);
>> +      else
>> +         vao->VertexBindingMask |= VERT_BIT(index);
>> +
>>         vao->NewArrays |= binding->_BoundArrays;
>>      }
>>   }
>> @@ -2126,6 +2131,8 @@ vertex_array_attrib_binding(struct gl_context *ctx,
>>
>>      assert(VERT_ATTRIB_GENERIC(attribIndex) < ARRAY_SIZE(vao->VertexAttrib));
>>
>> +   vao->VertexBindingMask |= VERT_BIT_GENERIC(attribIndex);
>
> So there are still a few problems here.  The _Enabled bitfield is a mask
> of vertex arrays, while VertexBindingMask is a mask of vertex buffer
> binding points.  There is no fixed coupling between these states, so the
> two bitfields can't be compared.  Yet valid_draw_indirect() compares them.
>
> bind_vertex_buffer() treats VertexBindingMask as a mask of vertex
> buffer bindings, while vertex_array_attrib_binding() treats it as a mask
> of vertex arrays.  The latter function also incorrectly assumes that if
> the association between an vertex attribute and a buffer binding point is
> changed, then the new binding point has a vertex buffer set.

There is perhaps some confusion with the name here, I agree that name of 
the mask is bad and suggested 'VertexAttribBufferMask' would be much 
better. Mask exists to check 'which vertex arrays have been associated 
with a vertex buffer', it is definitely not a 'mask of vertex buffer 
binding points' like you state above. My intention is that I'm setting 
those vertex array bits that have a vertex buffer and when drawing that 
needs to match which arrays are enabled.

Following that logic, bind_vertex_buffer() sets the bit for the vertex 
array which some buffer was bound for and in 
vertex_array_attrib_binding() we update bit of vertex array in the mask 
in same manner.

It is true that there is no check if such vbo exists, so if I understand 
you correctly this is what is actually missing?


> I suggest making VertexBindingMask a mask of vertex arrays instead,
> so it can be compared with _Enabled, and renaming it to
> VertexAttribBufferMask or something similar.  bind_vertex_buffer()
> then needs to be changed to set or clear the _BoundArrays bits in the
> mask instead of the index bit.  vertex_array_attrib_binding() also
> needs to check if the new binding has a buffer set, and update the
> attribIndex bit in the mask accordingly.
>
>>      vertex_attrib_binding(ctx, vao,
>>                            VERT_ATTRIB_GENERIC(attribIndex),
>>                            VERT_ATTRIB_GENERIC(bindingIndex));
>>

OK, I'll see how this would look. As it is legal to call 
glVertexAttribBinding with a binding index that does not have existing 
vbo I'm not sure if using BoundArrays makes this any better. The 
important part is to update the mask always when a change happens in 
binding.

// Tapani