Change-Id: I43c782a244996aac7bbafd9cb857a3063a68b5a1
Reviewed-on: https://swiftshader-review.googlesource.com/7391Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Bug swiftshader:6

Change-Id: Ib60cdecf28027687271d2456c6aae79065fb9804
Reviewed-on: https://swiftshader-review.googlesource.com/7364Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

Bug swiftshader:13

Change-Id: I054d0521ecbb1959c5c19b6fa54f76707f4e49c9
Reviewed-on: https://swiftshader-review.googlesource.com/7358Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Subzero has unrelated types for constants and variables. Both are
operands to instructions, but we can't express this relationship at
the Nucleus level. We'd require a large number of Value to Operand
conversions. Instead, an Assign instruction is provided to convert
a Constant into a Value.

Bug swiftshader:12

Change-Id: Ie35a2cea3e485c4012ed949f92825a41caca3367
Reviewed-on: https://swiftshader-review.googlesource.com/7370Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

This avoids having to store the type as a member field.

Bug swiftshader:12

Change-Id: I53ea4869a5b590027217d1e9cc015c0502ab530f
Reviewed-on: https://swiftshader-review.googlesource.com/7361Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

This refactoring paves the way for the next patch to support
constructing RValues from Constant pointers as well.

Bug swiftshader:12

Change-Id: I26df3da13314de556dc1af40388278d144560b97
Reviewed-on: https://swiftshader-review.googlesource.com/7371Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

"filter" option was ignored. Simply added the argument
to the context function to fix the issue.

Change-Id: I108c4128bbc4f319091f082df96e056057f926fe
Reviewed-on: https://swiftshader-review.googlesource.com/7363Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

When x was flipped, the entire row would be copied in the
reverse byte order, which was wrong for any type with pixels
larger than 1 byte. Now the full pixels are copied in the
reverse order instead of the bytes, so pixels are no longer
byte flipped.

Change-Id: Ie0e0516546a49d0f890a4abe24d44ccb3ffb2350
Reviewed-on: https://swiftshader-review.googlesource.com/7362Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

The code was not allowing to unset a sampler object from a texture,
so the faulty condition was removed to fix the issue.

Fixes 30 tests in:
dEQP.functional.samplers.*

Change-Id: Ibbc3389d3d6135db7bc5a2f87f0df5527b78c7fd
Reviewed-on: https://swiftshader-review.googlesource.com/7360Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Bug swiftshader:11

Change-Id: I794ef54a7c2ecde71ee6344c63955d2f838ff189
Reviewed-on: https://swiftshader-review.googlesource.com/7331Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Bug swiftshader:6

Change-Id: I1de2fe8983c7b8d29e0295e5b4705697e5dd6df1
Reviewed-on: https://swiftshader-review.googlesource.com/7292Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Change-Id: I5ab054a5622c9f9dee099f5a3a30c3bb4079ffeb
Reviewed-on: https://swiftshader-review.googlesource.com/7291Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Change-Id: I7eb124ce596d2c7e2419a01a995ea6fa871c8a88
Reviewed-on: https://swiftshader-review.googlesource.com/7290Reviewed-by: Nicolas Capens <capn@google.com>
Tested-by: Nicolas Capens <capn@google.com>

All LLVM instructions are LLVM values, but in the case of terminator
instructions their value is meaningless, and we don't use them.

Bug swiftshader:10

Change-Id: I9eb71e1dcea2614cb810021dfbf3c8908a7e2510
Reviewed-on: https://swiftshader-review.googlesource.com/7355Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

This removes all Common headers from Reactor headers, making it
easier to use Reactor outside of SwiftShader.

Bug swiftshader:10

Change-Id: I35c5d138e781d4e911d206954d38e392714d4659
Reviewed-on: https://swiftshader-review.googlesource.com/7330Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Bug swiftshader:10

Change-Id: I258fc7d66988c5bb9a20f98ac33e87ae26379855
Reviewed-on: https://swiftshader-review.googlesource.com/7280Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Just a refactoring, to make it clearer that Reactor types are the
higher level abstraction, with Nucleus the intermediate interface.

Bug swiftshader:10

Change-Id: I87d8a03cd05d7847511692c5ac9862e77d670801
Reviewed-on: https://swiftshader-review.googlesource.com/7279Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Bug swiftshader:10

Change-Id: I29b1de8c1adb67449a380c307d12e2aea21f32cc
Reviewed-on: https://swiftshader-review.googlesource.com/7251Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

It has bitrotted and should be reimplemented with support from the
Reactor back-end if needed again.

Bug swiftshader:10

Change-Id: Ie926903ccd8117e91b9d13bf031a7b88287e7276
Reviewed-on: https://swiftshader-review.googlesource.com/7250Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

It's only used for creating intrinsics calls, which is an
implementation detail.

Bug swiftshader:10

Change-Id: Ieac1303e08571e265e239e74c6ea84ece99b53e1
Reviewed-on: https://swiftshader-review.googlesource.com/7278Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Bug swiftshader:10

Change-Id: Ide80a6d97f67ae04d9118307984235f8f4c2e3ad
Reviewed-on: https://swiftshader-review.googlesource.com/7277Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Bug swiftshader:10

Change-Id: Iaed1c9a078b82516db67fe0de287e7ea0362dc03
Reviewed-on: https://swiftshader-review.googlesource.com/7276Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Bug swiftshader:10

Change-Id: Ib45d459aa4f68e572b1deaa03e0aaf83002b7881
Reviewed-on: https://swiftshader-review.googlesource.com/7275Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Bug swiftshader:10

Change-Id: I86b80cb03ff48ff7273aaf23a8e4995b3436f825
Reviewed-on: https://swiftshader-review.googlesource.com/7274Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Use only llvm::Constant in the interface instead.

Bug swiftshader:10

Change-Id: Ideb2ed1ee083e4e6896c4bc8e8830ea02972b487
Reviewed-on: https://swiftshader-review.googlesource.com/7273Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Bug swiftshader:10

Change-Id: I69105d8287e52f0f0573c45c59958994a56fc0af
Reviewed-on: https://swiftshader-review.googlesource.com/7214Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Bug swiftshader:10

Change-Id: I24fc66248bd2f66d2e8bc2b2a15e6c8fadc47efb
Reviewed-on: https://swiftshader-review.googlesource.com/7213Tested-by: Nicolas Capens <capn@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

"false" could not be converted to a pointer type. Changed it to nullptr.
Also changed a few other "return 0" to "return nullptr".

Change-Id: Idd4573d4627bb89f1b79e3b21de5af5bd6063916
Reviewed-on: https://swiftshader-review.googlesource.com/7357Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

Fixed first class array constructors by allowing basic type
arrays and structure arrays to be handled properly for the
EOpConstruct* operations.

This fixes all dEQP.functional.shaders.arrays.* tests.

Change-Id: I4fe99ec5256abf6483d3595890ba9c426abc97f8
Reviewed-on: https://swiftshader-review.googlesource.com/7351Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

There was an ambiguous type warning between Float4 and RValue<Float4>.

Change-Id: I526fc2430a66b8545e81fe5ae97fb423e7f01aec
Reviewed-on: https://swiftshader-review.googlesource.com/7356Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

It appears the Nucleus::createFPToUI function is broken and does
not yield the correct values for one pixel out of 4 when the f2u
operation is used. By manually rewriting the conversion using
signed int conversion and some arithmetic, the issue is solved.

This fixes at least 130 failures and also fixes many other tests
which were passing only due to the threshold being too lenient,
but which were showing obvious visual artefacts.

This affects, to the very least:
dEQP.functional.fbo.blit.conversion
dEQP.functional.shaders.functions.datatypes
dEQP.functional.shaders.operator.binary_operator
... and any other test using the f2u operation.

Change-Id: If38dad6b6ae8198f40e863d0847fa5080a2997e3
Reviewed-on: https://swiftshader-review.googlesource.com/7354Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

A few types were missing: 8b/16b, signed/unsigned integers

Change-Id: I1c168b5df941f6c01801406ba7879221b682fb96
Reviewed-on: https://swiftshader-review.googlesource.com/7353Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Fixes all failures in dEQP.functional.vertex_arrays.*

In OpenGL ES 3.0, vertex attributes streams can be of a
different type from the vertex attribute in the shader.
For this reason, some conversion may be required. This
cl solves this issue by:
1) Sending the information about the vertex attribute's
   type in the shader to the vertex routine.
2) Handling this information by adding conversion where
   appropriate.

Change-Id: I04a5a34aea12684209e584aa5f15a3edfd57f956
Reviewed-on: https://swiftshader-review.googlesource.com/7254Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

It appears that some bundling operations on Mac actually
require that all libraries have a unique name, since
output_dir is effectively ignored and overwritten in that
scenario, so I have to remove that renaming code on Mac so
that SwiftShader doesn't clash with Angle.

Change-Id: I3d36df525ad9229da0c35f27d367e7deef7e430d
Reviewed-on: https://swiftshader-review.googlesource.com/7271Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

To avoid duplicating code, Float4(RValue<Byte4>) and Float4(RValue<SByte4>) now call Int4(RValue<Byte4>) and Int4(RValue<SByte4>) respectively.

Change-Id: Ic49fdd151d15cc1c2753f535d3654028e7a1910a
Reviewed-on: https://swiftshader-review.googlesource.com/5560Tested-by: Meng-Lin Wu <marleymoo@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

It appears this issue might be fixable directly in chromium:
(https://codereview.chromium.org/2359913002/)
For now, removing the mac specific change.

BUG=31645700

Change-Id: I3d43c01f2d6500d67a298a047e8d98153f06dfb4
Reviewed-on: https://swiftshader-review.googlesource.com/7230Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

The following cl had broken all fragment uniform buffers:
https://swiftshader-review.googlesource.com/6950

Since the null pointer check was only made on the vertex
uniform buffer, it was preventing the fragment uniform
buffer from being set, which was causing a crash when this
happened.

There was a second issue  with this cl, which was allowing
'uniformBuffers[-1].get()' to be called, which should not
have been allowed.

Also, it is safer to call VertexProcessor::setUniformBuffer
and PixelProcessor::setUniformBuffer unconditionally in order
to set the null pointers correctly when no buffers are set.

Change-Id: I275b0bca6173ee3df4e0e369926ed0a94500e1ba
Reviewed-on: https://swiftshader-review.googlesource.com/7216Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Since the swiftshader libraries are loaded at runtime
by chromium, the correct way to add swiftshader as a
dependency is to use 'loadable_module' for mac and
'data_deps' for other platforms.

BUG=31645700

Change-Id: Ic393a4132fbfc44e71247fc0294cca2cb57d35c8
Reviewed-on: https://swiftshader-review.googlesource.com/7217Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

Tying up loose ends from the new output formats project.
One of the missing outputs was the RG8 format, which I added here.

Change-Id: I865ad4c462e892dbc3a7820003b2390b54453ee9
Reviewed-on: https://swiftshader-review.googlesource.com/7171Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>

There was a silly mistake in the code that I just found.
Instead of being ordered (1,2,3,4), the 4 pixels being
outputted were ordered (3,4,1,2), which was within the
tolerance threshold for most of the image, but exhibited
weird sparse diagonal lines of errors (values outside of
threshold). Reordered the pixels properly to fix the issue.

Change-Id: I082463d22e585e5ee98836a87ef60efca05e58f0
Reviewed-on: https://swiftshader-review.googlesource.com/7192Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <capn@google.com>