This optimizes evaluation of chained indexing operations by first
looking for an lvalue 'root' node and copying from it directly into the
topmost rvalue, instead of creating potentially many temporaries.

Change-Id: I47cf9b0230bd305695e7b2a44ddc44bc1320b764
Reviewed-on: https://swiftshader-review.googlesource.com/13908Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: I25251e94680624c8c85698e96a053f518bb2c367
Reviewed-on: https://swiftshader-review.googlesource.com/13968Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

After traversing the lvalue node tree we're fully done computing the
relative address and emitting any pointer arithmetic that might have
been needed. So the temporary that would have been used for that is no
longer needed.

Change-Id: I0a10bff979128f03544d0f8aa860c29f8867973a
Reviewed-on: https://swiftshader-review.googlesource.com/13948Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

This will enable reusing it to determine the root node of rvalues as
well. The only functional change is that struct indexing no longer
overrides the register type. This is of no effect here since lvalue
intermediates already inherited their type from the root node, but for
rvalues the intermediates are considered temporary registers, while
instead the root's type should be used.

Change-Id: I2dbd1b0f8886c3f111a2ed3ef7fe4e9a5b480085
Reviewed-on: https://swiftshader-review.googlesource.com/13930Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Previously we processed assignments in PostVisit, i.e. after both the
left and right side had been traversed. This produces temporaries for
the left hand side, which we don't use since we want to assign to the
lvalue. So instead we can explicitly traverse the right hand side, and
for the left hand side only traverse indirect indexing expressions.

Change-Id: I9ec0596a9c256921b65a9f70428d950959f66aa0
Reviewed-on: https://swiftshader-review.googlesource.com/13630Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

GLSL ES 3.0 interface blocks always have uniform storage qualification
and thus can't be used as l-value.

Change-Id: Iffe948dccbf2edb28eb1abe52a6f28dc76a1e84b
Reviewed-on: https://swiftshader-review.googlesource.com/13929Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

This matches Shader Model 4+ assembly syntax and helps debug shader
compilation issues.

Change-Id: Iff2a2991794e13476b99e11d6ac07fdf5d37570b
Reviewed-on: https://swiftshader-review.googlesource.com/13928Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

The integer modifier was used to emulate integer arithmetic using
floating-point operations, as was allowed/typical for OpenGL ES 2.0
implementations. Now that we support native 32-bit integer types and
we have separate opcodes for integer operations, it is not longer
needed.

Change-Id: I89987534c150d2426ac9f9e1e49b66f9deaee560
Reviewed-on: https://swiftshader-review.googlesource.com/13889Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Matrix elements cannot be accessed with dot notation.

Change-Id: I48258428e3fa5ffe9c95a8983516433b98ce233a
Reviewed-on: https://swiftshader-review.googlesource.com/13629Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

glDeleteSync can only delete an existing FenceSync object.

Change-Id: Ieb78894175235cc97c67ff0324d59bb5d7619918
Reviewed-on: https://swiftshader-review.googlesource.com/13829Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

GL_INVALID_OPERATION is generated if a non-zero vertex array object
is bound, zero is bound to the GL_ARRAY_BUFFER buffer object binding
point and the pointer argument is not NULL.

Change-Id: I48e73dca96bac2bd0496c202785e46e7d754dc11
Reviewed-on: https://swiftshader-review.googlesource.com/13830Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

While GL_ANY_SAMPLES_PASSED_CONSERVATIVE_EXT and
GL_ANY_SAMPLES_PASSED_EXT are mutually exclusive,
they can be used simultaneously with the
GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN query.

Change-Id: I818cf548c724ef56640b795530c3437e4dbcba98
Reviewed-on: https://swiftshader-review.googlesource.com/13828Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

- Can't update an already mapped bufferData
- Can't unmap an unmapped buffer
- Disallowed illegal mapBufferRange flag combinations
- Can't flush buffer if not mapped

Change-Id: I7013f63e5db64c1016f9ce50a43d0629f0a2950e
Reviewed-on: https://swiftshader-review.googlesource.com/13788Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

The dEQP tests for arccos/arcsin/arctan were failing due to
the precision of the arcsin/arccos/arctan approximation being
too low. It is possible to significantly improve this precision
with a few more terms, which is done here.

All arccos/arcsin/arctan dEQP tests pass.

Change-Id: I612551b48bdab03b81b2e72ddb8cd7f8502231c5
Reviewed-on: https://swiftshader-review.googlesource.com/13568Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Since we don't expose the ASTC related extension string,
we need to hide support for ASTC related enums. ASTC_SUPPORT
was added to easily re-enable ASTC, should we want to.

Change-Id: I3c65906c4729df8a2efb1a9a78e155bccdc715f9
Reviewed-on: https://swiftshader-review.googlesource.com/13749Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

It is an invalid operation to map an already mapped buffer.

Change-Id: Iba1c2389e5ef9c3114a2415279406ac7f08a0ed6
Reviewed-on: https://swiftshader-review.googlesource.com/13750Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

This implementation allows all dEQP precision test to pass
for sin, cos and tan function tests.

Change-Id: I33a24497dea68ab2de2e65931f50f2dd4298523c
Reviewed-on: https://swiftshader-review.googlesource.com/13555Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

Added proper checks for invalid operations.

Fixes all remaining failures in dEQP-GLES3.functional.lifetime*

Change-Id: Ia91ffdb8ced53f1f787b82ada0de22caf9ee9206
Reviewed-on: https://swiftshader-review.googlesource.com/13748Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Fixes all dEQP failures in inversesqrt and normalize.

Change-Id: Ifd1ac2714fa95bab1c720b65c95a27bbf23ea933
Reviewed-on: https://swiftshader-review.googlesource.com/13728Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

- Fixed how logarithm2() handles infinity
- Fixed a typo in exponential()

These fixes combined fix all dEQP failures for:
sinh, cosh, pow, exp, log, log2

Change-Id: I2d427892032fc5f75909a5609f606a787b08505a
Reviewed-on: https://swiftshader-review.googlesource.com/13708Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Added the missing implementation for isinf and isnan.

Change-Id: Ice603956a4ed8ef9515ae8f501b2876ce5fab584
Reviewed-on: https://swiftshader-review.googlesource.com/13608Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

We use icecc to speed up compilation in our project. After this change
https://swiftshader-review.googlesource.com/12308 chromium doesn't
compile via icecc as __has_include(<atomic>) is never expanded to 0 or 1
on icecc host and file <atomic> is not being included. My guess is that
icecc host gets partially expanded code and it's not being passed by
preprocessor again so condition #if __has_include(<atomic>) (expanded
from #if USE_STD_ATOMIC) is never fulfilled.
Bug: chromium:783135
Change-Id: I09cf120c399a794a15dec05d6edd74c7848fc53a
Reviewed-on: https://swiftshader-review.googlesource.com/13588Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Cezary Kulakowski <ckulakowski@opera.com>

Pass a surface to the blit/flip functions, instead of a raw pointer.
This puts the FrameBuffer in control of locking and unlocking.

Change-Id: I55335b3beef8d7083aae7687bd25392964261bde
Reviewed-on: https://swiftshader-review.googlesource.com/4482Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Previously we assumed the source stride (in pixels) to be the width
rounded up to the next multiple of 2. This is currently correct but
may not hold in the future.

Change-Id: I23a79ace66f720398a120b70081d731c2cf1b4c0
Reviewed-on: https://swiftshader-review.googlesource.com/4481Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

This mainly groups the state that is used for generating a new blit
routine into a second BlitState structure 'updateState'. It also
allowed for the FrameBuffer's own parameters to not have a 'dest'
prefix. Also, 'locked' was renamed to 'framebuffer', and 'target' to
'renderbuffer'.

Change-Id: I64e26f0b06f9f4419b8ca67e6fbb0dee8272898a
Reviewed-on: https://swiftshader-review.googlesource.com/11510Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

If both libGLES_CM and libGLESv2 are available, we were using the
former for creating the buffers of an EGL surface. The choice is
arbitrary, but it causes confusing issues when an interface changes
and one of the libraries wasn't rebuilt. This is less likely to occur
for libGLESv2.

Change-Id: If0c05d50a141c3782866892694d238332d8211db
Reviewed-on: https://swiftshader-review.googlesource.com/13388Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

The Reactor Value resulting from loading the value from the
dereferenced Pointer does not get passed to the Int4 cast constructor
correctly. This happens in 32-bit builds on Visual Studio
2017 (15.3.5), but not in 64-bit builds.

Change-Id: I68e1639ada436c65bb4f3bf9b3ac88d503f10e30
Reviewed-on: https://swiftshader-review.googlesource.com/13554Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Change-Id: I53b7938af132180dcf95312c76790fe761aab35d
Reviewed-on: https://swiftshader-review.googlesource.com/13553Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Cube texture sampling functions which take explicit gradients require
the gradients to divided by the same factor used to project the texture
coordinates onto the cube. Previously we assumed they were already in
the projected coordinate space.

Change-Id: I2825df17bee74b551b8d7491c050b7766653a65c
Reviewed-on: https://swiftshader-review.googlesource.com/13552Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

This is necessary for cube or 2D array shadow texture sampling
functions which need the fourth texture coordinate component for
depth comparison while also taking a lod or bias parameter.

Change-Id: I1e1399f134e22cecaff97a224df2c13c57ba3a40
Reviewed-on: https://swiftshader-review.googlesource.com/13551Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Change-Id: Ib62f310abecbc4cdaf6e9300791600f25af0eaf3
Reviewed-on: https://swiftshader-review.googlesource.com/13550Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Previously we dynamically allocated an array of SamplerCores. This
isn't necessary and we can just create one as a temporary object where
used. This has the added advantage that we could have Reactor variables
as class members and keep them short-lived.

Change-Id: Ifb2e6edbf275aa793bd7880bd35384e16000007d
Reviewed-on: https://swiftshader-review.googlesource.com/13548Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Depth format textures can have a comparison operation performed after
their texels have been sampled.

Change-Id: I49f6bb7fab9765265761144ee8b6b62439beb5a3
Reviewed-on: https://swiftshader-review.googlesource.com/5870Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Fixed glBindVertexArray. The early return was wrong,
binding vertex array 0 is perfectly valid.

All related dEQP tests pass.

Change-Id: Idd0ae745dc15877afb7c7323aab8253fb71e9017
Reviewed-on: https://swiftshader-review.googlesource.com/13549Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Varying variables can be structure in OpenGL ES3.
This cl adds support for structure varyings.

Change-Id: I4d1d80c6afed0a86a23b0a467d4764a4e08f133d
Reviewed-on: https://swiftshader-review.googlesource.com/13529Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

When multisampling is enabled, outline edges were getting
initialized to one of the  primitive's X position. If the
primitive was out of bounds, then the default position was
out of bounds, which led to an initial out of bounds memory
access.

Added a clamp to fix the issue.

Bug chromium:779364

Change-Id: I4661f4229ee28a3032c763ed18dde799d3c3926b
Reviewed-on: https://swiftshader-review.googlesource.com/13528Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

This will help ASan find bugs in the translator when fuzzing, by making
separate allocations instead of using suballocations within the pool.

BUG=swiftshader:86

Change-Id: Ic59f7eef2c7985f7bf545d41be956df478331e2d
Reviewed-on: https://swiftshader-review.googlesource.com/13488Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Corentin Wallez <cwallez@google.com>

Was an implicit narrowing from size_t (64-bit) to unsigned int (32-bit).

Change-Id: I9726d70370d97da1bae2652bbce9db929179a292
Reviewed-on: https://swiftshader-review.googlesource.com/13469Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Unused variable 'entry'.

Change-Id: I115b7d599d871c0f130f72e32618a8ed29776b81
Reviewed-on: https://swiftshader-review.googlesource.com/13468Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Bug swiftshader:86

Change-Id: If98cbf93e8b8d41246d9f06503da0752fe968ff8
Reviewed-on: https://swiftshader-review.googlesource.com/13448Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Corentin Wallez <cwallez@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>