Format validations were failing if the exact same format
wasn't used between different texture related calls to the
same texture, even when the formats were equivalent, so I
added a GetSizedInternalFormat function based on a format
map (courtesy of Angle). The validation checks were also
unified in utilities.cpp and used wherever texture formats
were used, to make sure Image objects used in Texture
objects always use the sized format, so that format
comparisons work properly.

Change-Id: I72fc8fb1b0f135ac679c274866e5b8e223541e7f
Reviewed-on: https://swiftshader-review.googlesource.com/4082Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Integer to integer types copy operations were failing due
to a missing assignment when calling the Insert function
and bad offsets on 16 and 32 bit component types.

Change-Id: I2db55e9b6a284f8349e8fef0451255ee33be5837
Reviewed-on: https://swiftshader-review.googlesource.com/4410Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Fixed a few mistakes in the stencil clearing code to fix both
masking and scissoring. Also added a few variable in both the
depth clearing code and stencil clearing code to make them
more readable.

Bug 25971463

Change-Id: If33e2b0685fd35ef2edd31c56aea088cfbf2b430
Reviewed-on: https://swiftshader-review.googlesource.com/4335Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Some types without alpha were not correctly setting the
alpha value, so copying to a type with alpha was causing
the image to not appear (due to it being almost completely
transparent).

Change-Id: Ib12c291bd5c51628e0579285731033372656f6a0
Reviewed-on: https://swiftshader-review.googlesource.com/4391Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

The new extensions are:
- GL_OES_framebuffer_object (with extra exports)
- GL_OES_depth24
- GL_OES_depth32
- GL_EXT_color_buffer_half_float

These were mostly supported already with the GLES3 related changes,
so only a few minor changes were required.

Change-Id: Iab6e94043cfdabb8ad74e526f44f7bd9d57c267f
Reviewed-on: https://swiftshader-review.googlesource.com/4342Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

We need this to define the native EGL types for Apple.

Change-Id: Iba492ea863fffd3649c8f7a3b7ca7453e12c3772
Reviewed-on: https://swiftshader-review.googlesource.com/4381Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Corentin Wallez <cwallez@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

- Implemented mipmap generation and completeness checks for
  Texture2DArray.
- Fixed texture parameters setters and getters along with
  mipmap generation functions to use the proper 2D array
  texture instead of the 3D texture for 2D array textures.
- Enabled the same path as 3D texture for 2D array textures
  in the sampler.
- Added an address function for the w component, which
  simply clamps the rounded value for 2D array textures and
  adapted SamplerCore::computeIndices to this new behavior.

Change-Id: Ida0659afac75330bfd9af4052cfd2625c729f9ef
Reviewed-on: https://swiftshader-review.googlesource.com/4310Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Change-Id: I83294568a5725e95fcf2d7490336bead35517bdf
Reviewed-on: https://swiftshader-review.googlesource.com/4362Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Greg Hartman <ghartman@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Framebuffer completeness, usually verified through
glCheckFramebufferStatus, should return
GL_FRAMEBUFFER_INCOMPLETE_MULTISAMPLE when the number of
samples is not the same for all attached renderbuffers.

Change-Id: I8d29895c6d4ded26847ce47fbd2581bbf06ed1bf
Reviewed-on: https://swiftshader-review.googlesource.com/4363Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Fixed a few things related to FramebufferTextureLayer:
- Added layer validation to make sure it does not exceed
  the size of the 3D texture's or 2D array's depth.
- Fixed frambuffer target / texture target confusion.
- Removed bad validation.
- Fixed last 2 arguments of set[insert type]Buffer calls
  (layer and level) which were inverted.

Change-Id: Ie1f2cb595d8b9abfee27bcf834c535f7e023e3ee
Reviewed-on: https://swiftshader-review.googlesource.com/4343Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Change-Id: I2bd7c62b0923c855327fdb105eb22ef4655ea8ac
Reviewed-on: https://swiftshader-review.googlesource.com/4344Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

Change-Id: Ibb2ebd66116f3dfd0008217153006bd6c7a49b9e
Reviewed-on: https://swiftshader-review.googlesource.com/4322Reviewed-by: Greg Hartman <ghartman@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

Change-Id: If348bb5dd92c1160e57f1db15929a2e02d1f253d
Reviewed-on: https://swiftshader-review.googlesource.com/4350Reviewed-by: Greg Hartman <ghartman@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

Change-Id: Ief54422c0c0d82c773f96f4491bfebb7ee717ba8
Reviewed-on: https://swiftshader-review.googlesource.com/4351Reviewed-by: Greg Hartman <ghartman@google.com>
Tested-by: Nicolas Capens <capn@google.com>
Reviewed-on: https://swiftshader-review.googlesource.com/4353Reviewed-by: Nicolas Capens <capn@google.com>

Change-Id: I06da8a8e48f202d6625c6d6eb8569be3867da13c
Reviewed-on: https://swiftshader-review.googlesource.com/4341Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>
Reviewed-on: https://swiftshader-review.googlesource.com/4352

BindRenderbuffer 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, but the same also applies
to GLES 2.0, which has been fixed here. Note that the Context
already handles the case where renderbuffer is zero properly.

Bug 25983901

Change-Id: I75f706fc56df28871bf60b5645f1c0db039fccb8
Reviewed-on: https://swiftshader-review.googlesource.com/4334Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

- Removed unused variables
- Fixed member initialization order in a few classes
- Fixed Surface::setSwapBehavior()
- Removed unused mPixelPackingStateDirty members
- Fixed initialization of "size" member in LinkedVarying class
- Fixed constness of a string
- Removed unused static functions
- Added parenthesis to fix && / || order ambiguity

Change-Id: Ia9ad8eaca335c60871fdc58037e441aa2010a641
Reviewed-on: https://swiftshader-review.googlesource.com/4301Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

A few minor fixes:
- In FramebufferRenderbuffer, if renderbuffer isn't 0,
  it must be a valid object.
- In FramebufferTexture2D, I moved the isCompressed check AFTER
  the textarget validation, to avoid using an invalid textarget.
- In GetFramebufferAttachmentParameteriv, not using GL_BACK,
  GL_DEPTH or GL_STENCIL for the default framebuffer should
  produce GL_INVALID_ENUM instead of GL_INVALID_OPERATION.
- In GetFramebufferAttachmentParameteriv, when
  attachmentObjectType is GL_NONE, in ES3, the query for
  GL_FRAMEBUFFER_ATTACHMENT_OBJECT_NAME should return 0.

Change-Id: I913cadd5961fa473b54ddfe174772bb7270dfdc5
Reviewed-on: https://swiftshader-review.googlesource.com/4333Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Since the new headers have been committed, a larger amount of
color attachments enums can be supported. Added the new enums
where appropriate.

Change-Id: Idd7210fbeaeb7828ca6c094bed115addd90b497c
Reviewed-on: https://swiftshader-review.googlesource.com/4332Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

This cl doesn't entirely enable integer texture sampling, but
it adds the required cases in SamplerCore. To fully enable
these types, only a few small selection functions will be
required so that the proper types use the proper sampling code.

Change-Id: Ie101d782dd9b43c96d8e0198c49fe9ec6855b007
Reviewed-on: https://swiftshader-review.googlesource.com/4150Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

This is odd, but stencil masking was simply disabled, so
clearing the stencil buffer would only work if the mask
was 0xFF, or it would simply skip the clearing entirely.
I removed this condition to fix the issue. Also removed
some dead code and added an early exit if the mask is 0.

Change-Id: I359b10ed3382b75cb9d078470f237e68f1a6e7b9
Reviewed-on: https://swiftshader-review.googlesource.com/4303Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

To make it easier to branch on the different texture
fetching options, a new TextureFunction class is
introduced here, which performs the string comparisons
and identifies the different options.

I also had to add a 5th argument for textureGradOffset
and textureProjGradOffset.

I added function stubs (with the UNIMPLEMENTED markers)
for all new texture functions.

Change-Id: I58cde91a2bacb0012bdc34ec85b0befa19a85326
Reviewed-on: https://swiftshader-review.googlesource.com/4116Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

According to the ES3 spec, there's no requirement for a
renderbuffer to be allocated until it is bound and any
non-zero renderbuffer bound must be allocated.

Change-Id: Id47083c7ec6e3b3698e176d6feff31121983e446
Reviewed-on: https://swiftshader-review.googlesource.com/4331Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Change-Id: I12fdf25465176ed9b0c2e95ef398b8b469419b6e
Reviewed-on: https://swiftshader-review.googlesource.com/4311Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

Change-Id: I9cff8a31cac858e14c2364a79f61c028996e91dd
Reviewed-on: https://swiftshader-review.googlesource.com/4304Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Greg Hartman <ghartman@google.com>
Reviewed-by: Keun Soo Yim <yim@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Change-Id: Ifff92974ea41be61244ae1c09e2d54e299c10eeb
Reviewed-on: https://swiftshader-review.googlesource.com/4160Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

Change-Id: I401c06e523d945d6e629c3326bb7cbb2054c48f1
Reviewed-on: https://swiftshader-review.googlesource.com/4307Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

Change-Id: I0790163f89116469f73ad9ff15c71dc7366a6f65
Reviewed-on: https://swiftshader-review.googlesource.com/4306Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

Change-Id: I68685932934caba4d9fab7b45014d24316ce45a8
Reviewed-on: https://swiftshader-review.googlesource.com/4293Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

Added both RGB8 and BGR8 formats to the blitter and related
functions so that these formats may be used with glReadPixels.

Change-Id: I22ee13f837b66af5f2135abc77fe81cc2e995fec
Reviewed-on: https://swiftshader-review.googlesource.com/4294Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

The "clear" operation can now be done through the blitter.
The few changes are:
- The blitter now supports RGBA masking
- The blitter now supports RGB565
- When in "clear" mode, the blitter does a one read/multiple writes

The old clearing code has been deleted from Surface.

Change-Id: I970c3a0323f63ee5c89f02d94a2705e4bcf83866
Reviewed-on: https://swiftshader-review.googlesource.com/4291Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

It is now possible to use floating point renderbuffers and read
the data back from them. The changes include:
- Modified glReadPixels so that it always uses the blitter to
  copy the data to the external buffer.
- Added new types to both Framebuffer and some utility functions.
- Added the new ValidReadPixelsFormatType function to validate the
  glReadPixels format/type combo, which had a bit more
  possibilities than the RGBA/UNSIGNED BYTE combo previously used.

Change-Id: I1726ea57c4f7aa85bf0ffa7f323dc6a16abc34ff
Reviewed-on: https://swiftshader-review.googlesource.com/4260Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Fragment outputs weren't given a proper index, so a
fragment output array has been added to fix this.

Change-Id: Id9be21f60cc23528bf1af078a8b3ca2df28e7ee4
Reviewed-on: https://swiftshader-review.googlesource.com/4295Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Added some extra constructors that simply truncate from a type
which has a higher number of bits. The new constructors are:
- Byte from UInt
- Byte from UShort
- SByte from Int
- SByte from Short
- UShort from UInt

Also added an implementation of the RoundUInt function using the
UInt from Float constructor, which had to be fixed since it was
using createFPToSI instead of createFPToUI.

Change-Id: Ie7ee21ef20fbb8133b9f7c74afa1fec9e6c51957
Reviewed-on: https://swiftshader-review.googlesource.com/4300Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Conversion from a float format to a non float format was
broken in the blitter because the clamp operation, either
to the [0,1] range or the [-1,1] range, was happening
after the scaling operation, so non float types were
becoming either -1, 0 or 1, which was wrong.

Change-Id: I3e1290313043fc49030454916b6e4ea6666f8343
Reviewed-on: https://swiftshader-review.googlesource.com/4290Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Implemented the missing pieces for multiple draw buffers support:
- Fixed Context::getScissoredImage() to use the drawbuffer
  parameter properly
- Enabled setting multiple render targets
- Added dynamic indexing of gl_FragData using a new dynOut flag
  to figure out which channels are being written to.

Change-Id: Id3d95c46a980a698f71e99f7781cc6287e880e9d
Reviewed-on: https://swiftshader-review.googlesource.com/4220Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Change-Id: I51c1774a35706639481c030b4365ee2e5d7ccd4d
Reviewed-on: https://swiftshader-review.googlesource.com/4255Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

RGB_565 is now as fast if not faster than 32-bit formats.

Change-Id: I6358e35557164541601ac76ae11028afba439084
Reviewed-on: https://swiftshader-review.googlesource.com/4253Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

crbug/546671

Change-Id: Ie43f8e34fa22bb0d23e4a123e789c429f2024ebb
Reviewed-on: https://swiftshader-review.googlesource.com/4254Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

This forces clang to avoid use init_array for all initialization rather than
splitting the initialization between init_array and ctors.

Local builds do this anyway, but for some reason the builds on go/ab do not.
Setting this flag makes the builds consistent (and unbroken on JB-MR1).

I already checked this in on cloud-android-current-release because I wanted
to test it before submitting.

Bug 25597090

Change-Id: I5a7195ba53531835fb8333b698ed39d21ff847a6
(cherry picked from commit 95c7182f118cc686df27c3f49db7d3c7908e0e59)
Reviewed-on: https://swiftshader-review.googlesource.com/4271Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Greg Hartman <ghartman@google.com>