- Updated preprocessor code from Angle revision
  9fc8733187c02470a9eadb6d295348b6d37a2004
- Reran generate_parser.sh (flex 2.6.4, bison (GNU Bison) 3.0.4)
- Made a few trivial changes in src/OpenGL/compiler in order to
  adapt to the new preprocessor code.

Fixes all 24 failures in:
dEQP-GLES3.functional.shaders.preprocessor.*

Change-Id: I00d0b511d617ab81a0f57310174e1ba8bf7c22e5
Reviewed-on: https://swiftshader-review.googlesource.com/15109Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

The ETC2 decoder spec has a slight difference with the EAC decoder
spec when it comes to handling the 0 multiplier corner case.

For ETC2, we have (OpenGL ES 3.0 spec, section C.1.3):
"An encoder is not allowed to produce a multiplier of zero, but
 the decoder should still be able to handle also this case (and
 produce 0 x modifier = 0 in that case)."

For EAC, we have (OpenGL ES 3.0 spec, section C.1.5):
"If the multiplier value is zero, we should set the multiplier
 to 1.0/8.0"

In order to take this into account, the EAC decoded output can no
longer be represented by an 8 bit value, but must be represented by
a minimum of 11 bits, as the spec requires. For now, the EAC decoder
decodes EAC into a 32 bit integer format, which then gets converted
to a 32 bit float format internally.

Eventually, it would be possible for the EAC decoder to decode the
image to a signed 16 bit integer internal format, if it was supported.

Fixes all failures in:
dEQP-GLES3.functional.texture.wrap*

Change-Id: I32106383ade56e375229231ff230a2574791caa6
Reviewed-on: https://swiftshader-review.googlesource.com/15188Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: I879eb3c3d9093f769f88bd3a2dcc67f91bd6eb47
Reviewed-on: https://swiftshader-review.googlesource.com/15049Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Fixes (suppresses) Visual Studio 2017 warning (error) about std::tr1.

Change-Id: I3319dfa77d00e81e48ef7ef0ab884e131e9a039e
Reviewed-on: https://swiftshader-review.googlesource.com/15048Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

This stores the layer as part of the framebuffer object, instead of the
renderbuffer, and passes it to the renderer at draw time.

Change-Id: I947f8a58ba7a119cc9d890659c99e7e05d0d1d91
Reviewed-on: https://swiftshader-review.googlesource.com/14728Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

This adds the ability to render to a specific layer of a 2D array
texture.

Change-Id: I3055d62c9964b2729704785efb068b82eef4f9ce
Reviewed-on: https://swiftshader-review.googlesource.com/14668Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Previously, multisampled images used the 'depth' member of 3D images
or 2D arrays as the number of samples. This caused rendering to a
layer of a 2D array to be interpreted as rendering to a multisampled
render target. This change adds a 'samples' member which is orthogonal
to 'depth'.

Note that write operations put the same color into each of the samples,
while read operations (still) assume multisampled images have been
resolved into the first slice.

Change-Id: Ib33a0cf8194e19fcbb569b0c257ba1e1bd9c4821
Reviewed-on: https://swiftshader-review.googlesource.com/14808Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

The stencil buffer always has a quad layout, and can't be copied using
copyBuffer() even when copying the full buffer, because it copies one
row at a time (i.e. it assumes a linear layout).

Likewise a depth buffer may optionally have a quad layout and copying
the full buffer still doesn't allow using copyBuffer().

To keep maximum performance for the typical case of copying between
equally sized buffers, a new code path it added to copy the entire
slice in one memcpy() call.

Change-Id: I77f60f66e5cdcd7cf0a73d03f3d3485809d7ebe6
Reviewed-on: https://swiftshader-review.googlesource.com/15008Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Several depth-only formats were implemented using depth+stencil
formats. This resulted in unintended stencil operations happening, as
well as unnecessary allocations and blitting of stencil data.

This change also introduces quad-layout depth+stencil formats for more
efficient rendering.

Note that as before the 'internal' buffer's format also indicates the
presence of a stencil component, even though it only stores the depth
component, while the 'stencil' buffer now has an S8 format or NULL for
a depth-only format.

Change-Id: I245f0cb5a999851e24082f3ab1ea78a5f5956af3
Reviewed-on: https://swiftshader-review.googlesource.com/14988Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

drawBuffer and readBuffer default to GL_BACK in the default
Framebuffer, but for all other Framebuffer objects, the
default value should have been GL_COLOR_ATTACHMENT0.

Fixes all failures in dEQP-GLES3.functional.state_query.integers*

Change-Id: I060db67c50561d2678d57de1d7067bd7f8d2a53a
Reviewed-on: https://swiftshader-review.googlesource.com/14948Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Added pureInteger member to VertexAttribute in order
to keep track of whether the attrib was set using
glVertexAttribIPointer or glVertexAttribPointer and
properly return that state when querying
GL_VERTEX_ATTRIB_ARRAY_INTEGER.

Fixes dEQP-GLES3.functional.state_query.shader.vertex_attrib_integer

Change-Id: Ie6cfcd2008f7abb61d457a41124600fe7cea229a
Reviewed-on: https://swiftshader-review.googlesource.com/14828Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

The typeSpecifier.array check was doubled inside and outside of
the mShaderVersion check. The typeSpecifier.array check should
only be done when mShaderVersion < 300.

Fixes dEQP-GLES3.functional.state_query.shader.program_active_uniform_types

Change-Id: I2907676c346a9f53d4f8c8da172b7ce0bc7b4083
Reviewed-on: https://swiftshader-review.googlesource.com/14848Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

VertexAttribute objects should have a default size of 4.

Fixes dEQP-GLES3.functional.state_query.shader.vertex_attrib_size

Change-Id: I597d2095f7be087f35a4adfb01bb2a3ba5200701
Reviewed-on: https://swiftshader-review.googlesource.com/14928Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

If the format and internalformat parameters are the same, only the
combinations in Table 3.3 of the OpenGL ES 3.0 specification are valid.

GL_RGB10_A2 and GL_RGB10_A2UI have GL_UNSIGNED_INT_2_10_10_10_REV type.
GL_OES_vertex_type_10_10_10_2 is a vertex attribute type, part of the
GL_OES_vertex_type_10_10_10_2 extension.

GL_RGB10_A2 internal format is valid for glReadPixels with a
combination of format GL_RGBA and type GL_UNSIGNED_INT_2_10_10_10_REV.

Change-Id: I590b43fcf9f1dc4beee9a64b45fe94fd74388b7a
Reviewed-on: https://swiftshader-review.googlesource.com/14788Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

The read buffer state is part of the framebuffer object, so determining
the index is moved to the Framebuffer class. Also make sure GL_NONE is
handled without causing invalid access by using GL_INVALID_INDEX.

Change-Id: I26476024cd8dd820b0e3f53d9c3dd5a7d46c082e
Reviewed-on: https://swiftshader-review.googlesource.com/14748Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: I62047cd0bf6c9f066e7a63c1316288d227caa9aa
Reviewed-on: https://swiftshader-review.googlesource.com/14768Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Bug: 70191319
Test: $ lunch aosp_arm-userdebug; BOARD_VNDK_VERSION=current m -j
      $ emulator # boot to home screen

Change-Id: Id19a847072f2f0b55184c183647889c1a1b53aef
Reviewed-on: https://swiftshader-review.googlesource.com/14688Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

In order to avoid creating confusion between shader model and
the actual shader's version, as specified by the #version
token in a glsl shader, version was renamed to shaderModel.

Change-Id: I481b6e3fc43168b504c2f2d9506422a697abc3a2
Reviewed-on: https://swiftshader-review.googlesource.com/14629Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Device::stretchRect() now supports out of bounds coordinates.
To avoid linear interpolation errors, source coordinates now
have to be in floating point rather than integer format. Most
changes in this cl are just to accommodate that int->float
change for the source rect.

Fixes all (28) failures in:
dEQP-GLES3.functional.fbo.blit.rect

Change-Id: I8fd017e60b61f2d7d6517b0e648b324be441cddd
Reviewed-on: https://swiftshader-review.googlesource.com/14648Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Used range-based for loop where it was trivial to do so.
This change should be noop in terms of functionality.

Change-Id: I3d692cc2706f35f5b710e7539fa084365cf28af1
Reviewed-on: https://swiftshader-review.googlesource.com/14628Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: I3a239325ed802e9188839c1c743d45930643c273
Reviewed-on: https://swiftshader-review.googlesource.com/14608Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Renderbuffer objects are defined to have an initial internalformat of
GL_RGBA4, but the red/green/blue/alpha bits are 0. This inconsistency
is resolved by setting the internalformat to GL_NONE but reporting
GL_RGBA4 when queried.

Change-Id: Ie9a342c05eaa23f81773b37ebb942ca2e5b1addb
Reviewed-on: https://swiftshader-review.googlesource.com/14588Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

GL_BACK, GL_DEPTH, and GL_STENCIL are only valid for the default
framebuffer, while GL_DEPTH_ATTACHMENT, GL_STENCIL_ATTACHMENT,
GL_DEPTH_STENCIL_ATTACHMENT, and GL_COLOR_ATTACHMENTi are only valid
for non-default framebuffer objects.

Also implement the color encoding query.

Change-Id: I153ae9407850a30ed14d9ae145ee3504ba71029a
Reviewed-on: https://swiftshader-review.googlesource.com/14569Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

- Attribute location aliasing was allowed prior to shader
  version 300, so location aliasing is now possible.
- Attribute binding refers to the linked location of attributes,
  so locations set using glBindAttribLocation() will only be
  returned by glGetAttribLocation() after the program is linked.
  Before that, it will return the location allocated during the
  previous glLinkProgram() call.

In order to do that, an extra map was added.
"linkedAttributeLocation" represents the attributes' location,
as a result of linking a program.
"attributeBinding" represents the attributes' future location,
when the next program linking occurs.

On top of that, the shader's version was not being passed down
from TranslatorASM to es2::Shader, or from es2::Shader to Program,
so this information also needed to be properly transferred.

Fixes all failures in:
dEQP-GLES3.functional.attribute_location*

Change-Id: I4ba7dc7c2f6d444e805cadeb5445f5ff371c3d95
Reviewed-on: https://swiftshader-review.googlesource.com/14568Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Textures and renderbuffers were storing both the OpenGL internalformat
and the SwiftShader implementation format (i.e. sw::Format). This
change removes the implementation format, only keeping it in the Image
class.

Change-Id: Ie6ac96f6450b9a55ea9b49c6cf4b2c0681e95522
Reviewed-on: https://swiftshader-review.googlesource.com/14528Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: Ib153f6af75c982fae0325a104da3c0a4fc9ee9dc
Reviewed-on: https://swiftshader-review.googlesource.com/14548Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

- Use internalformat parameter if valid, instead of deriving from
  format/type parameters.
- Validate format/type/internalformat parameters in CopyTexSubImage().
- Moved early-out optimizations after validation.
- Removed duplicate validation.
- Use GLint consistently for internalformat parameters.

Change-Id: I377c6bb5381602d13d281f19985aa4f11d201099
Reviewed-on: https://swiftshader-review.googlesource.com/14488Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

- Added *ALPHA8*/*LUMINANCE8* formats and *_SNORM formats
  to non color renderable formats
- Added a new function IsMipMappable in order to allow
  mipmapping of the *ALPHA8*/*LUMINANCE8* formats

Fixes all (24) failures in:
dEQP-GLES3.functional.texture.mipmap.2d.generate*
dEQP-GLES3.functional.texture.mipmap.cube.generate*

Also fixes WebGL test:
conformance/textures/misc/texture-npot.html

Change-Id: I5f3210094fbc5b2e5bae25c88a5ef4f1ffb69cbd
Reviewed-on: https://swiftshader-review.googlesource.com/14130Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Because of the 3 reserved uniforms used for gl_DepthRange, the
number of uniforms allowed in fragment and vertex shaders was
too small by 3, so their allowed number was increased by 3.

As for attributes and varyings, their numbers needed to be
increased in order to pass all OpenGL ES 3.0 dEQP tests.

Change-Id: I6527334e1503c08281303915654087cbac8089a0
Reviewed-on: https://swiftshader-review.googlesource.com/4034Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: I29e493f91df9ebef948b02096421102eb328d545
Reviewed-on: https://swiftshader-review.googlesource.com/14490Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Drawing without a current program is not an error and we can early-out,
but not until after other validation which could generate an error has
completed. Validation happening on a valid program happens afterwards
though. Likewise, providing insufficient vertices for even one
primitive results in drawing nothing, but validation still needs to
happen.

Change-Id: I5385ffe352fc38343caa41eb99f5549472da3b4f
Reviewed-on: https://swiftshader-review.googlesource.com/14489Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Added some validation for uniforms and uniform blocks.

Fixes some failures in (only 1 failure left):
dEQP-GLES3.functional.negative_api.state*

Change-Id: I507c7e8784230f85b0d312d162e5ff82bdbd15ed
Reviewed-on: https://swiftshader-review.googlesource.com/14428Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: I03a369139b733a75a7790e65afc514b929e32090
Reviewed-on: https://swiftshader-review.googlesource.com/14468Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Bug chromium:769026

Change-Id: I059117a0209e3ac13877736abafe3599310f0670
Reviewed-on: https://swiftshader-review.googlesource.com/14448Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

The addressing is offset by 1 to account for the border. Note that this
could be avoided by locking at (0, 0) instead of (-1, -1) instead, but
then negative address offsets have to be allowed and this complicates
the subsequent calculations and only unsigned extension from 32-bit to
64-bit is typically for free during memory accesses.

Change-Id: I5fb843401d440a9d77d141782124a9c260765830
Reviewed-on: https://swiftshader-review.googlesource.com/13289Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

The border addressing mode should ideally blend with the border color
when using linear filtering. Until we have a border of pixels around
2D textures, we should probably avoid bleeding pixels from the opposite
edge and just clamp the coordinates.

Change-Id: I7afbb629998732b62413e00ba7bcd55340c7b9bb
Reviewed-on: https://swiftshader-review.googlesource.com/14289Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Change-Id: Ic259645ab7950b0b2800964bbfd14f3294de50b1
Reviewed-on: https://swiftshader-review.googlesource.com/14288Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

The cube texture borders will be used for linear interpolation, in
order to produce seamless edges.

Change-Id: Idd17c72c6aaf7dcc65188b065ac8ba179b58cc37
Reviewed-on: https://swiftshader-review.googlesource.com/8209Tested-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Borders are required to support seamless cubemap sampling. Subsequent
patches will fill the borders with pixels from adjacent cube faces.

The border is expressed in pixels and is added on all edges,
resulting in an image of dimensions
(width + 2 * border) x (height + 2 * border).
The surface still exposes dimensions of width x height through the API
and points to the same pixel when locked.

Change-Id: I06d5121267ce1a2c50e628490d8690de71bfeb08
Reviewed-on: https://swiftshader-review.googlesource.com/8208Tested-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Uses 0, -1 offsets instead of 0, 1 to be able to use it as a
mask. Optimize calculation of the integer coordinates and fraction by
always having xyz1 = xyz0 + 1 (before wrap). Optimize the
clamping/wrapping of the integer coordinates. Skip addressing operation
of the third component for cube sampling (since already projected to
2D face).

Change-Id: If2e7c74aac9ae923a0c1ffc278fcdfec00f216f3
Reviewed-on: https://swiftshader-review.googlesource.com/14269Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>