Added fuchsia-specific versions of allocateExecutable() and
markExecutable().

Bug: chromium: 778467
Change-Id: I027f1409f7b0343e102f98fcc34f2e93c2d626e4
Reviewed-on: https://swiftshader-review.googlesource.com/c/23408Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Sergey Ulanov <sergeyu@chromium.org>

llvm::parseIR in llvm-7.0\llvm\lib\IRReader\IRReader.cpp calls parseAssembly
which is defined in lib/AsmParser/Parser.cpp, but the latter file is not
included in swiftshader_llvm which causes unresovled symbol for linking.
This CL added the necessary source files under llvm\AsmParser to
swiftshader_llvm.

This CL also changed __rdtsc() to use Windows ARM64 intrinsic _ReadStatusReg
to get cycle counter as alternative of __rdtsc().

Bug: chromium:893460
Change-Id: I269662f2e4249a3ec5495ecad02bc759139e1d4f
Reviewed-on: https://swiftshader-review.googlesource.com/c/23508Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Tom Tan <Tom.Tan@microsoft.com>

The .def file is already in sources, which means GN will already add
this flag automatically (cf `gn help sources`).

No intended behavior change.

Bug: none
Change-Id: Ic8204bbdd9da47f3731d48e62ccfeb880ac9eddc
Reviewed-on: https://swiftshader-review.googlesource.com/c/23528Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nico Weber <thakis@chromium.org>

Change-Id: I37dfecd8c92f3f34c117432956d169f4b191c849
Reviewed-on: https://swiftshader-review.googlesource.com/c/23248Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Milko Leporis <milko.leporis@mips.com>

Blitting from a higher bit depth source to a lower bit depth
destination was not working properly because Vulkan expects
clamping, while the blitter was truncating the values.

From section 19.5. Image Copies with Scaling:
"Signed and unsigned integers are converted by first
 clamping to the representable range of the destination
 format, then casting the value."

This fixes all remaining failures in:
dEQP-VK.api.copy_and_blit.core.blit_image.all_formats.color.*

Bug b/119620767

Change-Id: Ibd1c1e4be4f5107e7d423781b46021f4c5df1070
Reviewed-on: https://swiftshader-review.googlesource.com/c/23368Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Added support for reading the following formats in the Blitter:
- VK_FORMAT_B10G11R11_UFLOAT_PACK32
- VK_FORMAT_E5B9G9R9_UFLOAT_PACK32

Fixes all failures in:
[dEQP-VK.api.copy_and_blit.core.blit_image.all_formats.color]
.b10g11r11_ufloat_pack32.*
.e5b9g9r9_ufloat_pack32.*

Bug b/119620767

Change-Id: Ie5824b59b0193115cc7d142efc41cc3ac516c05c
Reviewed-on: https://swiftshader-review.googlesource.com/c/23348Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

Implemented Float <-> Half conversion. The conversion is copied from
Half.cpp and is the same as the C++ code, converted to Reactor code.
The conversion is NOT optimized. Optimizing this code path will be
done later, this cl only cares about correctness.

Passes all conversion tests to and from half float in the following
patterns:

[dEQP-VK.api.copy_and_blit.core.blit_image]
.all_formats.color.[supported format].r16_sfloat.*
.all_formats.color.[supported format].r16g16_sfloat.*
.all_formats.color.[supported format].r16g16b16a16_sfloat.*
.all_formats.color.r16_sfloat.*
.all_formats.color.r16g16_sfloat.*
.all_formats.color.r16g16b16a16_sfloat.*

Bug b/119620767

Change-Id: Icd725dcd2ddd8c54e6657a2c3210da270210a149
Reviewed-on: https://swiftshader-review.googlesource.com/c/23329Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Added support for VK_FORMAT_A1R5G5B5_UNORM_PACK16 in the Blitter.

Passes all tests in:
[dEQP-VK.api.copy_and_blit.core.blit_image.all_formats.color]
.b4g4r4a4_unorm_pack16.a1r5g5b5_unorm_pack16.*
.r8g8_snorm.a1r5g5b5_unorm_pack16.*
.a8b8g8r8_unorm_pack32.a1r5g5b5_unorm_pack16.*
.a8b8g8r8_snorm_pack32.a1r5g5b5_unorm_pack16.*

Bug b/119620767

Change-Id: I366cb5697b60b4f0fb51931a6be7b8732cf789ec
Reviewed-on: https://swiftshader-review.googlesource.com/c/23328Reviewed-by: Corentin Wallez <cwallez@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

Added enums for mostly implemented format, along
with the implementation for 2 new formats:
VK_FORMAT_B4G4R4A4_UNORM_PACK16
VK_FORMAT_A2B10G10R10_UINT_PACK32 (as a true integer format)

Bug b/119620767

Change-Id: I139408609cdbc35f7d9b0a8220e86a598c70a9e9
Reviewed-on: https://swiftshader-review.googlesource.com/c/23309Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Corentin Wallez <cwallez@google.com>

Added missing VkFormat entries to the following functions:
Surface::hasQuadLayout
Surface::isFloatFormat
Surface::isUnsignedComponent
Surface::isSRGBformat
Surface::isCompressed
Surface::isSignedNonNormalizedInteger
Surface::isUnsignedNonNormalizedInteger
Surface::componentCount
Surface::selectInternalFormat

Bug b/118429780

Change-Id: I46f8fb3f8a14f82a2b6615e129e4de0443666eca
Reviewed-on: https://swiftshader-review.googlesource.com/c/23308Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Corentin Wallez <cwallez@google.com>

First simple implementation of image blitting.
Note that layers are not yet taken into account.

Passes almost all tests in (There are only 2 failures,
which both have an off by 1 error on a single pixel):
dEQP-VK.api.copy_and_blit.core.blit_image.simple_tests.*

Bug b/118619338 b/119620767

Change-Id: I1e0ac88089d6159924569099ea6345804a219d2c
Reviewed-on: https://swiftshader-review.googlesource.com/c/23268Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Added buffer copy command and support for VkBufferCopy regions.

Passes all tests in:
dEQP-VK.api.copy_and_blit.core.buffer_to_buffer

Bug b/118619338

Change-Id: Id69cd3f089b07343a7209290808c11d4c17f1a49
Reviewed-on: https://swiftshader-review.googlesource.com/c/23229Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Added support for image regions and buffer rowLength/imageHeight
for Image to Buffer copies or Buffer to Image copies.

Passes all tests in:
dEQP-VK.api.copy_and_blit.core.image_to_buffer
dEQP-VK.api.copy_and_blit.core.buffer_to_image

Bug b/119620767

Change-Id: I2b2592c06d52609c3f9c51fa164a0335f58f361f
Reviewed-on: https://swiftshader-review.googlesource.com/c/23228Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Added implementation for Depth and Stencil which currently allocates
Depth and Stencil as 2 contiguous image planes in memory.

Copying to and from depth and stencil planes is now supported which
fixes all failures in:
dEQP-VK.api.copy_and_blit.core.image_to_image.all_formats.depth_stencil
Which were the last remaining failures in:
dEQP-VK.api.copy_and_blit.core.image_to_image

Bug b/119620767

Change-Id: Ie90737c219c70d9c8a68c024ed33292812a3b7f4
Reviewed-on: https://swiftshader-review.googlesource.com/c/23209Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Added basic support for array layers in Image objects.
This allows allocating the memory for multiple layers,
copying between a 2D image with multiple layers to and
from a 3D image.

Passes all tests in:
dEQP-VK.api.copy_and_blit.core.image_to_image.3d_images.*

Bug b/119620767

Change-Id: I305650ba89f90dcd527b484fc3ed095a309506d1
Reviewed-on: https://swiftshader-review.googlesource.com/c/23208Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

This fixes a memory leak in the Queue object. The sw::Renderer
object has a non trivial destructor, which means we need to call
it before deallocating the Queue objects.

Bug b/117974925

Change-Id: I1e13c6108e77d5cdcbea337d572f72fdd32530a1
Reviewed-on: https://swiftshader-review.googlesource.com/c/23188Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Change-Id: I5feee7fb928a49762faa8154804fc4b0e86f60b8
Reviewed-on: https://swiftshader-review.googlesource.com/c/23169Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

Piped all the information from the vertex buffers and
the pipeline to the renderer to perform a draw.

In order for the renderer and the context to have proper
lifetimes, they both reside in the Queue object for now.

Bug b/118619338

Change-Id: Ifa03acd13ceb065a856b50f2cffadd4ee6b9a163
Reviewed-on: https://swiftshader-review.googlesource.com/c/23111Reviewed-by: Corentin Wallez <cwallez@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

Implemented vkCmdBeginRenderPass and vkCmdEndRenderPass.
The RenderPass object begin and end are noop for now, but
the framebuffer gets cleared within vkCmdBeginRenderPass.

Bug b/119620965 b/118619338

Change-Id: I8e1f2932d9d52b3dcb5c10d4e60cba83b28bb95d
Reviewed-on: https://swiftshader-review.googlesource.com/c/23108Reviewed-by: Corentin Wallez <cwallez@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

Adjusted a few checks in the GraphicsPipeline constructor,
specifically stages' specialization info and sample mask,
which can either be null or default.

Also fixed destroying Pipeline objects, making sure there's
no confusion between GraphicsPipeline and ComputePipeline's
destroy() methods inherited from their 2 base classes
(Pipeline::destroy() and ObjectBase::destroy()).

Bug b/118386749

Change-Id: I78be9f0bbb735518e07d52f7372dbf08b9834abf
Reviewed-on: https://swiftshader-review.googlesource.com/c/23110Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Use the blitter to implement Image::clear.

The image can be easily packed into a sw::Surface. For now,
the surface is a temporary, but it may become a member in
the future if it makes sense.

Bug b/119620767

Change-Id: I122a321509dc8cc5b624ddd776bd9059eee0cbc4
Reviewed-on: https://swiftshader-review.googlesource.com/c/23109Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Added VkImageView object.
Added "clear" function to illustrate the link between VkFramebuffer
attachments, VkImageView and VkImage, along with the minimal set of
class members to allows the function arguments to be used properly.

Bug b/119620767 b/119621736

Change-Id: I0e7e6017979960bacbdf888d632c83edda3483cf
Reviewed-on: https://swiftshader-review.googlesource.com/c/22668Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Implemented Image to Image copy, Image to Buffer copy and
Buffer to Image copy commands which are required in order
to run the dEQP-VK.api.copy_and_blit.core.image_to_image.*
tests.

Passes over 99.7% of the 10907 tests in:
dEQP-VK.api.copy_and_blit.core.image_to_image.*

Bug b/118619338 b/119620767

Change-Id: Id4d3aa5186bed84d0925e4424399716efbd241ee
Reviewed-on: https://swiftshader-review.googlesource.com/c/23071Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Corentin Wallez <cwallez@google.com>

A call to vkBeginCommandBuffer implicitly resets the command buffer.

Bug b/118619338

Change-Id: I020a2d15915cbb309dbcd012b2dbbf34ca2f63c6
Reviewed-on: https://swiftshader-review.googlesource.com/c/23148Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Corentin Wallez <cwallez@google.com>

Image objects now have the ability to copy their content to another
image object. They can also copy their content to or from Buffer
objects.

Bug b\119620767

Change-Id: I047e9ecdcc11e264589de1d9461ef448f22a4d9e
Reviewed-on: https://swiftshader-review.googlesource.com/c/23070Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

Includes only the include/spirv/unified1 subtree.

Based on git://github.com/KhronosGroup/spirv-tools revision 17da9f8231f78cf519b4958c2229463a63ead9e2

Bug: b/120799499
Change-Id: Ifb790bf95035d7feb2d54d07e1475013dfd298b1
Reviewed-on: https://swiftshader-review.googlesource.com/c/23048Tested-by: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Corentin Wallez <cwallez@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Ideally we'd use the [[fallthrough]] attribute to silence this, but that
isn't supported until C++17 and we're sticking with C++11 for now.

Bug b/120382288

Change-Id: I98f28e42505ed496b0b35c29a7dc03fca18cf4ff
Reviewed-on: https://swiftshader-review.googlesource.com/c/23128Reviewed-by: Greg Hartman <ghartman@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

Buffer now has the ability to copy its content to or from memory.

Bug b/118383648

Change-Id: Ic01bf049fe054559299830ca750263c2277e4d86
Reviewed-on: https://swiftshader-review.googlesource.com/c/23069Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

For the purpose of running dEQP tests, for now, fences can be
implemented as noop. Once more complex rendering tests exercise
this, we'll be able to implement and test fences.

Bug b/117835459

Change-Id: I2b54be816c83f7a626108f61bd14eb9c969a2b0f
Reviewed-on: https://swiftshader-review.googlesource.com/c/23068Reviewed-by: Chris Forbes <chrisforbes@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

Because of existing synchronization mechanisms in SwiftShader,
pipeline barriers can be noop for now. If we do remove those
synchronization mechanisms, there's also a simple way of
implementing pipeline barriers, which is commented on here.

Bug b/118619338

Change-Id: If82d275c46f234e5549018115c9f19f21e2c09b2
Reviewed-on: https://swiftshader-review.googlesource.com/c/22708Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Chris Forbes <chrisforbes@google.com>

One part of this might change -- master contains some uses of
std::make_unique which Alexis might get rid of.

Change-Id: I40ae17f4e88920c4d60b79e1d4b117a7800fcd82
Reviewed-on: https://swiftshader-review.googlesource.com/c/23028Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Chris Forbes <chrisforbes@google.com>

MacOS' linker dislikes local static variables that are used
in two separate libraries. Removing them from the header files
fixes all the warnings.

Bug b/chromium:907088

Change-Id: I7b8ed44bf9a3180489a7407980740fd3f3863046
Reviewed-on: https://swiftshader-review.googlesource.com/c/22889Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

This cl adds the code needed to record and submit the minimal
subset of commands required to run a simple triangle example.

The commands themselves are still unimplemented.

Bug b/116336664

Change-Id: Id0109980225a64a2bb3599a89a5495091926c635
Reviewed-on: https://swiftshader-review.googlesource.com/c/22168Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Corentin Wallez <cwallez@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Added a few class members and adjusted the VkPipeline constructor
to extract basic parameters which will be used for rendering, like:
- Basic sw::Context parameters
- Scissor
- Viewport
- BlendConstants

Note: sw::Context should eventually be replaced by VkPipeline and be
      used directly by sw::Renderer once all existing OpenGL ES 3.0
      has been converted from VkPipeline.

(Chris: rebase + build system fixes for CMake path)

Change-Id: I7e7a9ff3c768da8e8c1e9461e140af4986429602
Reviewed-on: https://swiftshader-review.googlesource.com/c/22613Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Corentin Wallez <cwallez@google.com>

Change-Id: I45349cd6530814efa0c466beb0df7d53b4e17948
Reviewed-on: https://swiftshader-review.googlesource.com/c/23008Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Chris Forbes <chrisforbes@google.com>

Bug b/114402930

Change-Id: I228222003e5f015e4966eacc32094a48a7054d0f
Reviewed-on: https://swiftshader-review.googlesource.com/c/22988Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Kevin Schoedel <kpschoedel@google.com>

Added functionality so that the CommandPool can allocate and
free command buffer objects.

Bug b/119827933

Change-Id: I190ba3cd7738f2b5e37b196a0abba6d07cfae173
Reviewed-on: https://swiftshader-review.googlesource.com/c/22748Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

We basically replace all references to CMAKE_SOURCE_DIR and
CMAKE_BINARY_DIR with CMAKE_CURRENT_SOURCE_DIR and
CMAKE_CURRENT_BINARY_DIR.

This enables other cmake projects to take a direct dependency on this
project.

Test: cmake . still shows error for in source builds.
Test: mkdir build && cd build && cmake .. -G Ninja && ninja succeeds
Change-Id: Ibb8366f8e7fd3b3152634c71324a5182d269e647
Reviewed-on: https://swiftshader-review.googlesource.com/c/22692Reviewed-by: Lingfeng Yang <lfy@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Erwin Jansen <jansene@google.com>
Tested-by: Erwin Jansen <jansene@google.com>

The Subzero JIT only supports constants in the second operand of
arithmetic operations, not the first. It assumes constant folding
already took place (true for NaCl which takes LLVM IR as input).
Reactor has constant folding as part of the static C++ compilation,
but the Optimizer may substitute the first operand for a constant
(i.e. constant propagation).

Addressing it in the Optimizer by not performing the substitution
is not trivial because we'd have to check each use. And it would
cost run-time performance. Ideally we'd have a pass for
legalization/optimization which performs constant folding.

For now, avoid hitting 'unreachable' code in Subzero by multiplying
constants at the Reactor level.

Fixes regression caused by https://swiftshader-review.googlesource.com/22910

Bug chromium:904265
Bug swiftshader:14

Change-Id: Ifdc72ac997ad5d71c1f7006259f54f3d715cb613
Reviewed-on: https://swiftshader-review.googlesource.com/c/22930Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

GLSL textureSize() calls that use a sampler that is not a uniform,
should not use the index of the register as the texture unit index.
Instead the index is the value stored in the register itself.

Bug chromium:904265

Change-Id: Iee1058e789daae15248ae5ffa940a0155dca61b5
Reviewed-on: https://swiftshader-review.googlesource.com/c/22910Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>