Change-Id: I491411ba77addcb514944717b19918e4d1b98898
Reviewed-on: https://swiftshader-review.googlesource.com/16228Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

The surface set during bindTexImage should be unset during
releaseTexImage. Failing to do so may keep a dangling pointer
to the surface, which can cause a crash later on (this was hit
during the testing of new unit tests which  will be added soon).

Change-Id: Ic258fff8c880d449add825c5e9185435cc4ea419
Reviewed-on: https://swiftshader-review.googlesource.com/17548Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

The spec mentions them as color-renderable, but our IsColorRenderable()
function only handles sized internal format, and adding them there
might hide bugs in our renderbuffer implementation since they can only
be constructed using sized internal formats.

dEQP also has a note about this inconsistency, and consequently these
formats are not part of the 'mustpass' set for
dEQP-GLES3.functional.state_query.internal_format.
So it might be a spec oversight.

Change-Id: I08ca3dad4c0b5686536dafd3ade1e855e3968aa0
Reviewed-on: https://swiftshader-review.googlesource.com/16410Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Current build system assumes single arch targets to be 32bit and produces 64bit libraries
in an incorrect folder.

The change allows build system to determine proper location of the libraries.

Change-Id: Ic95a36f7108fcae8e6f120f0853ee00e9a8926c1
Reviewed-on: https://swiftshader-review.googlesource.com/17188Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

GL_APPLE_texture_format_BGRA8888 specifies that when using the BGRA_EXT
format, it "must have an RGBA internal format". Note that
GL_EXT_texture_format_BGRA8888 also allows BGRA_EXT as internalformat.

Change-Id: I27e032054240efe96c32e27ebbdfb3c28213b1d8
Reviewed-on: https://swiftshader-review.googlesource.com/16248Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

We were expecting R or RG internal formats, instead of L, A, or LA.

Change-Id: I6ee145341819460dda262b017db6d8105860f49a
Reviewed-on: https://swiftshader-review.googlesource.com/17508Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Imported texture rectangle tests from angle and fixed all
texture rectangle related failures:
- Rectangle textures can be rendered to
- Rectangle textures only support level 0
- Rectangle textures can't be compressed
- glTexStorage2D can no longer create a texture larger
  than the maximum size allowed

Change-Id: I089291c94aad79e244782a8d56dd224c7510d237
Reviewed-on: https://swiftshader-review.googlesource.com/16908Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

This makes it easier to use a different address,
in particular the port.

Also add a missing header for std::max.

Change-Id: I454dd6f4a92b5d8b97625e4cb50796e9f0940009
Reviewed-on: https://swiftshader-review.googlesource.com/16668Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

glBindRenderbufferOES() doesn't require anything except a
non-zero renderbuffer to create a new Renderbuffer object (it is not
required to use GenRenderbuffers before using BindRenderbuffer).
This had been fixed for GLES 3.0 and 2.0, but the same also applies
to GLES 1.1. Note that the Context already properly handles the case
where renderbuffer is zero.

Bug b/25983901

Change-Id: If5e97be8027fca7c97cce5c44d063cd054443700
Reviewed-on: https://swiftshader-review.googlesource.com/17490Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

When sampling from an incomplete texture, the OpenGL spec states that
(0, 0, 0, 1) will be returned to the shader from the texture lookup
function. Swizzling is only described for filtered texture values.

Change-Id: Ie8a02b3b0b9e571fd2f86302e777a4ff985da022
Reviewed-on: https://swiftshader-review.googlesource.com/17488Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

With the egl::Image class now only storing the sized internal format,
some OpenGL ES 1.1 functionality broke because it was still expecting a
base format.

Change-Id: Ib41e91f371a06b2a09471ea45dd1f8f56b94091f
Reviewed-on: https://swiftshader-review.googlesource.com/17468Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

GL_NONE is primarily a renderbuffer format, and GL_NO_ERROR more
clearly conveys the meaning when used as a validation return code.
Also, GL_NONE is not defined in OpenGL ES 1.1 (but GL_NONE_OES is,
as a renderbuffer format).

Change-Id: Ie1c00009c30c8d2717de15c7e27db295ecd9de1e
Reviewed-on: https://swiftshader-review.googlesource.com/17489Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Surface::resolve() currently only supports these formats. Also this
change reuses the GetColorComponentType() utility function instead of
using a separate switch. Effectively, signed normalized formats now
return 0 samples.

Change-Id: I839ffa89a63d8e1b0a40913d3b1ad13e2ce0b5a8
Reviewed-on: https://swiftshader-review.googlesource.com/16389Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

We weren't handling several of the format/type/internalformat combos
from table 3.2 of the OpenGL ES 3.0.5 spec. In particular those where
the format/type of a glTexSubImage2D() call can be used to update
images with an internal format not directly corresponding to it. Some
of these cases were handled using the blitter, but not all GL formats
have a SwiftShader equivalent. Also, the blitter is slower than
specialized C++ pixel transfer code, and the blitter's fallback path is
even slower.

This patch provides specialized pixel rectangle transfer code for each
combination of formats. We also now only store the effective sized
internal format of the images. Validation also happens using the sized
internal format wherever feasible, instead of unsized formats or
SwiftShader formats.

Change-Id: Id55db490002ab8fc2f16f766c43b43f121e5768e
Reviewed-on: https://swiftshader-review.googlesource.com/17429Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

While in general one should strive to never hit an assert, and thus we
should not worry about what happens in release mode when the condition
is false, there are cases in which it becomes very hard to prove that a
condition will always be true. For instance when another layer is
responsible for validation, but the code experiences a lot of change,
it could inadvertently expose an untested bad edge case which is only
detected much later. For these cases, it is wise to still perform a
less harmful action than just continue.

The ASSERT_OR_RETURN macro will cause a return from the calling
function when the condition is false.

Bug b/73656151

Change-Id: I36953bb5c477ecca67647ae1a7e2fb97e7e4fca6
Reviewed-on: https://swiftshader-review.googlesource.com/17428Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

When the external buffer of a surface is dirty, and we're trying to
lock the internal buffer (or vice-versa), an 'update' needs to happen.
This resulted in us locking the internal buffer for writing, thereby
marking it as dirty. This triggered an assert which checks that both
buffers can't be dirty at the same time. Also, in release mode this
could result in redundant updates when the external buffer is locked
again.

We should mark the destination buffer dirty when we're about to make
sure it's up to date, so a new lock type was added which is equivalent
to a write lock but without setting the dirty flag.

Also, we were allocating memory for a 0x0 stencil buffer for each
render target, when there's no stencil component. So return nullptr
when the format is NULL.

Change-Id: Ie7b5528e3eedc3c3efdf8461047e6284b7bdfc84
Reviewed-on: https://swiftshader-review.googlesource.com/16828Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Neither GL_EXT_texture_format_BGRA8888 nor
GL_APPLE_texture_format_BGRA8888 make mention of GL_BGRA_EXT
or GL_BGRA8_EXT being accepted as the internalformat parameter of
glCopyTexImage2D, but there's a reasonable assumption that textures
with BGRA format can be used as the read color buffer.
GL_EXT_texture_format_BGRA8888 does mention that the format is
color-renderable, and GL_APPLE_texture_format_BGRA8888 adds it to table
3.2 (but fails to mention changes to table 3.16).

Bug swiftshader:96

Change-Id: I6d18faeaded73164a2c6af87ae6109fc77825964
Reviewed-on: https://swiftshader-review.googlesource.com/17448Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Fixes building SubzeroTest on Mac.

Change-Id: I8ca1f2ea8101bec496b8e9741c38025ff932b3d6
Reviewed-on: https://swiftshader-review.googlesource.com/17189Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: I0c186ddd43c28a3f2c49321da23d5db44abe9dbf
Reviewed-on: https://swiftshader-review.googlesource.com/17148Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Added missing resource initialization in order to prevent getting
"uninitialized value" errors when running the fuzzer.

Change-Id: I342441832a159b79cbf036d7ce87f05fcaf15e8c
Reviewed-on: https://swiftshader-review.googlesource.com/17128Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Change-Id: If8871d9286ef7c4d09fd075ad5c7b707079158dd
Reviewed-on: https://swiftshader-review.googlesource.com/17108Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

TParseContext::addConstVectorNode() was returning a
TIntermTyped* variable of the wrong type if a swizzle
was of a different size than the vector's size. When
parseVectorFields() fails, fields.num is reset to 1,
which no longer matches fieldString.size(), so
fieldString.size() should not have been used. To
avoid future errors, the proper type is now set
directly within TParseContext::addConstVectorNode().

Bug chromium:796776

Change-Id: Id5e8d1b94d0a03fcf5ef6f5deeb77581bc070288
Reviewed-on: https://swiftshader-review.googlesource.com/17089Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: Ida0a270c1e0ab0900109b7d3bce90dfee9a01960
Reviewed-on: https://swiftshader-review.googlesource.com/17091Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: I53def31f6e63a9cf0884715078bbf236f00f3c51
Reviewed-on: https://swiftshader-review.googlesource.com/17090Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Apply xoffset, yoffset, and zoffset at image buffer lock so we don't
have to pass them down to the row loading functions as parameters.

Change-Id: I055549c80b1b207df4c49f6f59ecfb8286736040
Reviewed-on: https://swiftshader-review.googlesource.com/17068Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

glCompressedTexSubImage2D() takes an internalformat parameter which
is already always a sized internal format.

Change-Id: Iaf783e019e08d941053d5f347ee495bdd179f1ce
Reviewed-on: https://swiftshader-review.googlesource.com/17050Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

We were passing the framebuffer to the texture's copying method, where
the first render target was used, instead of the read buffer.

Change-Id: I344b4082f675b7e5fc6a73af31a6764ab4d3b6a3
Reviewed-on: https://swiftshader-review.googlesource.com/17049Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Allow not specifying a fragment output location.

Change-Id: I76c6e6efdff3161586f5d5422e97731969b9396c
Reviewed-on: https://swiftshader-review.googlesource.com/17028Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

We were enumerating the color buffer formats by sized internal format,
which is simplified and made less fragile by retrieving the base
internal format and implementing the spec's validation tables more
directly.

Change-Id: I46d244b9ed738e927b0cb5fe6d1deebf6f0cbae5
Reviewed-on: https://swiftshader-review.googlesource.com/16988Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: Iddd32408b090f00cbae15a76cab064d699f5098b
Reviewed-on: https://swiftshader-review.googlesource.com/16968Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Also remove default constructors and unused members.

Bug chromium:801648

Change-Id: I822ca1e1569708ca661796ee9252bae68a0a284a
Reviewed-on: https://swiftshader-review.googlesource.com/16948Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

This refactoring makes the code for texture image upload significantly
smaller. It also aligns better with the layout of table 3.2 in the
OpenGL ES 3.0 specification which lists valid format/type combinations.

Custom RGB565 format handling was removed because it corresponds with a
regular 2-byte per pixel memcpy.

Sized internal formats were removed because a recent change fixed not
using the internalformat parameter as upload format.

Change-Id: Ief6c6e60c3d9a734cac241022bc95ffc88d752c8
Reviewed-on: https://swiftshader-review.googlesource.com/16568Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

We were using the internal format to compute the size of the pixels
provided for texture image upload. They can differ in size from the
format/type combination.

Change-Id: I7f3cd535cdefefc3c7ec4695d35d181b43ae20a4
Reviewed-on: https://swiftshader-review.googlesource.com/16629Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Deduplicate common functions and remove unsized color buffer formats.

Change-Id: I76b3b507c27c8bfa932aab64a9338e2a124a25da
Reviewed-on: https://swiftshader-review.googlesource.com/16289Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

We were generating GL_INVALID_OPERATION when a non-compressed format
is used where a compressed one is expected or a compressed one is used
where a non-compressed one is expected. Instead we can generate
GL_INVALID_ENUM since these enums are never valid for these calls
(GL_INVALID_OPERATION is for when the enum is valid in some situations
but it's currently used incorrectly).

Change-Id: I9ee92a92cda0574284e709a5a745d5bb8b825a31
Reviewed-on: https://swiftshader-review.googlesource.com/16888Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

The previous fix wasn't handling the case where the same fragment
output variable would go through declareFragmentOutput() multiple
times, so this cl fixes it.

Fixes all dEQP-GLES3.functional.shaders.random failures.
Still fixes original WebGL2 conformance tests failures in:
all/deqp/functional/gles3/fragmentoutput

Change-Id: Ia9c4f5ed5c444ab0c020cac8be511fcaad23c55d
Reviewed-on: https://swiftshader-review.googlesource.com/16928Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Unsigned integers in GLSL were being parsed using the regular
integer parser, so it was limited to INT_MAX. All values from
INT_MAX + 1 to UINT_MAX could not be parsed properly.

Also, added constant folding for the 4 bit conversion glsl
functions.

Fixes shader compilation issue in the Epic Zen Garden example:
https://s3.amazonaws.com/mozilla-games/ZenGarden/EpicZenGarden.html
(unfortunately, the screen is still black, so there are other
 issues left)

Fixes WebGL 2 test: conformance2/glsl3/float-parsing.html

Change-Id: Iae52b2c8e083f0e1a22599e5a583297b9850444d
Reviewed-on: https://swiftshader-review.googlesource.com/16648Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: Ifd2c846c274840f19c1ea8c75fda42bf5640caa4
Reviewed-on: https://swiftshader-review.googlesource.com/16808Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

SwiftShader wasn't correctly handling having no restart indices
in the index buffer while having primitive restart enabled. In
the WebGL 2 conformance tests, the reference image is also
computed while primitive restart is on, which means that
Context::drawElements() may modify the internal primitive type,
so indices must be adjusted to reflect that change, even if no
restart indices are present in the index buffer.

Fixes all WebGL 2 tests in:
all/deqp/functional/gles3/primitiverestart

Change-Id: I24f129e0f7bd42b5037b74617b48a32621440ddb
Reviewed-on: https://swiftshader-review.googlesource.com/16548Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

GL_MAX_RECTANGLE_TEXTURE_SIZE_ARB was getting queried by Chromium
since GL_ARB_texture_rectangle is exposed, and if this query fails,
WebGL does not get enabled.

Change-Id: Ie6b40f35f0cfc4547079e303609afa3778010dd1
Reviewed-on: https://swiftshader-review.googlesource.com/16849Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>