xref: /aosp_15_r20/external/deqp/doc/testspecs/GLES31/functional.tessellation.txt (revision 35238bce31c2a825756842865a792f8cf7f89930)

-------------------------------------------------------------------------
drawElements Quality Program Test Specification
-----------------------------------------------

Copyright 2014 The Android Open Source Project

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

     http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
-------------------------------------------------------------------------
    Basic tessellation tests

Tests:
 + dEQP-GLES31.functional.tessellation.*

Includes:
 + Tessellation UVW coordinate verification
   - Coordinates read with transform feedback (with point_mode)
   - Triangles, quads, isolines
   - equal_spacing
   - fractional_odd_spacing and fractional_even_spacing with tessellation
     levels chosen so that results are not implementation-dependent
 + Separate verification of fractional spacing modes
 + Winding modes (cw, ccw)
 + Built-in inputs and outputs and basic data transfer between TCS and TES
   - Read gl_PrimitiveID, gl_PatchVerticesIn in both TCS and TES
   - Read gl_TessLevelInner[] and gl_TessLevelOuter[] in TES
   - Differing input and output patch sizes
   - Pass data in (gl_in/gl_out []) .gl_Position between VS, TCS and TES
   - Basic barrier usage
 + Rendering result sanity verification
   - Check for no obvious gaps or overlaps in tessellation triangulation for
     quads and triangles modes
   - Isolines rendering
 + Test for absence of cracks at shared edges of primitives
   - Use either the 'precise' qualifier or carefully chosen primitive
     orientations
 + Test that patches are discarded when appropriate
 + Invariance rule testing
 + Passing per-patch and per-vertex simple and aggregate user-defined data
   from TCS to TES
   - Arrays, structs and interface blocks
   - Omit vertex IO array size expression, or use gl_MaxPatchVertices or
     an integer literal (queried from GL_MAX_PATCH_VERTICES)

Excludes:
 + Complete checking of validity of the filling triangles in the triangles and
   quads modes
 + Negative testing

Description:

The tessellation test group contains various cases testing basic tessellation
shader functionality.

The tessellation coordinate (gl_TessCoord) verification cases draw the vertices
of a tessellated "triangles", "quads" or "isolines" shape using the
"point_mode" input layout qualifier in the tessellation evaluation shader (TES).
The vertices are read with transform feedback, and the results are compared to a
point set generated by a reference tessellation vertex generator, verifying that
for each point in the reference set, there is an approximately equal point in
the GL-generated set, and vice versa. There is a small difference allowed in all
comparisons, and thus not all rules are exactly verified (such as
x + (1.0-x) == 1.0 for a given uvw coordinate x). The cases draw multiple times
with different tessellation levels.

The tests for the fractional spacing modes (fractional_odd_spacing,
fractional_even_spacing) draw multiple simple isolines (with point_mode), each
resulting in a row of vertices. These vertices are read with transform feedback
and their x coordinates are inspected to see whether the edge is split into
valid segments. The conditions for the amount, length and location of the two
"additional segments", as laid out in the specification, are checked. The
segment sets produced by different isolines are also compared to each other to
determine whether the implementation fulfills the requirement of monotonic
segment length growth, and the requirement that two edges with identical clamped
tessellation levels have identically-located segments.

The winding mode cases draw triangles or quads with a specific winding mode
("cw" or "ccw") input layout qualifier in the TES. Back face culling is enabled.
Each case draws two times, with glFrontFace(GL_CCW) and glFrontFace(GL_CW), and
verifies that the primitive is drawn iff the glFrontFace value matches the
winding mode specified in the TES. The patterns drawn are simple and verified by
counting background-colored and non-background-colored pixels.

The cases testing gl_PrimitiveID and gl_PatchVerticesIn read the variable in
either TCS or TES; the TCS cases pass the value as a "patch out" output to the
TES. For the gl_PrimitiveID cases, multiple primitives are drawn and one of
them, with gl_PrimitiveID matching a specific value, is drawn with a differing
color; for the gl_PatchVerticesIn cases, just one primitive is drawn and the
value of gl_PatchVerticesIn is verified. The "quads" mode is used.

The gl_TessLevelInner[] and gl_TessLevelOuter[] cases simply verify that reading
the value of a tessellation level in the TES gives the value set in the TCS. The
arrays are indexed with literals. The "quads" mode is used.

The cases testing differing input and output patch sizes draw a simple shape
using an output patch size either less or greater than the input patch size. The
"quads" mode is used.

The barrier case uses both per-patch and per-vertex TCS output variables,
doing simple assignments and reads with them in different barrier()-separated
stages. The "quads" mode is used.

For the above-mentioned cases, the gl_PrimitiveID cases, the barrier case, and
the cases with differing input and output patch sizes are verified with
comparisons to pre-computed reference images; the gl_PatchVerticesIn and
gl_TessLevel*[] cases draw simple all-white patterns and are verified as such.

The basic rendering cases for the "triangles" and "quads" modes contain two
types of cases: cases testing that there are no obvious gaps in the
triangle-filled area, and cases testing that there are no overlaps between
triangles of differing inner layers of the tessellated primitive. The gap cases
draw a single primitive deformed so as to reveal possible errors in the
triangulation. The overlap cases draw a single primitive with different colors
for different concentric triangles (for the "triangles" mode) or concentric
rectangles (for the "quads" mode); the coloring is done so as to make it visible
if a fill-triangle spans between inappropriate inner triangles or rectangles.
Basic cases verifying the rendering results of isolines also exist; the cases
draw curly isolines with different colors for different vertices.
The above cases are repeated for different spacing modes. The results of the
rendering cases are compared to pre-computed reference images.

The shared-edge cases draw multiple primitives (with the "triangles" or "quads"
mode) that have shared edges, and check that no cracks appear at the edges. For
each pair of primitives sharing an edge, that edge has the same tessellation
level for both primitives. The primitives are deformed in the TES so that
potential cracks are likely to become visible. The resulting image is verified
by checking that no pixel (inside a fixed rectangle) remains uncovered by a
primitive. Two variants exist, with different methods for avoiding cracks:
either the 'precise' qualifier is used in the tessellation shaders, or input
patches are generated in a way that causes each shared vertex to have the same
index in all the primitives it belongs to.

The patch-discarding tests draw multiple patches in one draw call, with
different combinations for tessellation levels; some of these patches contain
relevant outer tessellation levels with a negative or zero value; they should
be discarded. Verification is done firstly by a quick check that the number of
vertices returned by transform feedback is correct, and secondly by checking
that white primitives exist in the area where non-discarded primitives are
drawn, and only there.

The invariance tests target the 8 invariance rules laid out in the
specification. They render one or more primitives with various tessellation
levels and/or programs with different configurations, read the primitives with
transform feedback, and do appropriate comparisons or other validation.

The tessellation coordinate generation is somewhat implementation-dependent when
using a fractional spacing mode and a tessellation level that isn't already
properly rounded. In many of the cases using a spacing mode other than
equal_spacing, the tessellation levels are chosen in a way that makes the
results non-implementation-dependent.

The exact manner in which a "triangles" or "quads" primitive is divided into
triangles is also implementation-dependent and is not thoroughly tested.

More complete checking of final triangle generation are possible candidates for
future tests.