- 12 Dec, 2019 5 commits
-
-
Sergey Ulanov authored
LLVMReactor was using default memory mapper when creating SectionMemoryManager. The default memory mapper calls mmap() to allocate memory on all platforms except Windows. It doesn't specify MAP_JIT flag that is required to create executable memory on Mac and Fuchsia. As result the allocated memory is not executable, which leads to crashes when trying to run the generated code. Update LLVMRector to use a custom MemoryMapper implementation. Bug: chromium:1032622, b/145348318 Change-Id: Id188af269d80371b709e625c29b92ff546e9ba28 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39408 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
Sergey Ulanov <sergeyu@chromium.org> Reviewed-by:
Nicolas Capens <nicolascapens@google.com>
-
Alexis Hetu authored
There was a missing prepareForSampling call after doing an image copy between 2 compressed textures. This cl fixes these angle end to end tests: CopyCompressedTextureTest.Basic/ES3_Vulkan_SwiftShader CopyCompressedTextureTest.Basic/ES2_Vulkan_SwiftShader Bug: b/146052572 b/146157683 Change-Id: I8703fa71557509e9ecb0d52a4ad2da4c72fa6812 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39448 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Reviewed-by:
Alexis Hétu <sugoi@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
-
Nicolas Capens authored
The expression "IfThenElse(e < 24, mantissa >> e, Int(0))" gets rewritten into "IfThenElse(e<24, mantissa> > e, Int(0))" by some versions of clang-format, thus breaking the build. It's likely a clang-format bug that should be fixed, but the workaround is simple and still quite elegant: "IfThenElse(e < 24, (mantissa >> e), Int(0))" Bug: b/144825072 Change-Id: I9cc19c6cae31bb0452b4e5402f6a742d15c45e55 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39428 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Reviewed-by:
Ben Clayton <bclayton@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Alexis Hetu authored
Added a block compression (BCn) decoder and logic for all formats for n in the [1, 5] range. The decompression follows the same logic as the ETC2 decompression in vk::Image. Tests: dEQP-VK.*bc* Bug: b/146052572 Change-Id: I64fac0a7af52e1be209c1cfd5373744918c7df14 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39369 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Ben Clayton authored
This block declares variables, so should be scoped. Bug: b/145351270 Change-Id: I90b750a81f4ff93f1c618a2ae969fa321961673c Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39311Tested-by:
-
- 11 Dec, 2019 10 commits
-
-
Ben Clayton authored
Server implements a Debug Adapter Protocol server that listens on a specific port, that operates on the vk::dbg::Context passed to the constructor. Bug: b/145351270 Change-Id: If09d92c6d298cfbd71a7b2366b6b8407a728ec43 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38899Reviewed-by:
-
Ben Clayton authored
Holds the global debugger state. Bug: b/145351270 Change-Id: I4e483c145af068755716f23c09adcc68d138ceaa Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38898Reviewed-by:
-
Ben Clayton authored
This CL contains minor tweaks to the style of our code to make clang-format produce prettier code. Bug: b/144825072 Change-Id: I821ecd3b9f408a4034992c2cc7ece2ef8552fe08 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39314Tested-by:
-
Antonio Maiorano authored
Bug: swiftshader:142 Change-Id: I188fb7895f5f52c4d5eb85ca5a4fc091ee95b604 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39368Reviewed-by:
Antonio Maiorano <amaiorano@google.com>
-
Ben Clayton authored
`err` was getting nullified when the program terminated abnormally, preventing the regex matching logic from being called. Should fix binning of abortion caused by SwiftShader debug macros. Change-Id: I811441296841c77cb87839039cde754d13f2a108 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39309 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
Paul Thomson <paulthomson@google.com> Reviewed-by:
-
Nicolas Capens authored
This is a style change. Visual Studio defaults to indenting namespace contents, and this was adopted for a long time, but with the new Vulkan implementation this was abandoned. However the legacy code borrowed from the OpenGL ES implementation still used indentation so it was inconsistent. The justification for not indenting namespace contents is that namespaces are merely a way to avoid name clashes with other projects we don't control directly (and in rare cases internal subprojects when we want to reuse the same names). Hence the vast majority of files have a single namespace, and unlike indentation used for ease of discerning control flow blocks, class contents, or function contents, which can become highly nested, there is no such readability advantage to indenting namespace contents. This is also the Google style recommendation (though no justification or discussion is provided): https://google.github.io/styleguide/cppguide.html#Namespace_Formatting One reasonable counter-argument is consistency with other blocks of curly brackets, but considering that most namespaces span almost the entire file, it's a substantial waste of line length. Because there is no indentation, there's also no need to have the open and closing brackets line up as a visual aid, like we prefer for other uses of curly brackets. So we place the open bracket on the same line as the namespace keyword. A comment is added to the closing bracket to discern it from other closing brackets. It also makes it easier to find the end of anonymous namespaces which typically go at the top of the source file. This change is make separately from applying clang-format because diff tools mark all these unindented lines as changes and this makes it hard to review the smaller style changes made by clang-format. The OpenGL ES and Direct3D code is left untouched because it is in maintenance mode and in case of regressions we want easy 'blame' tool usage. Bug: b/144825072 Change-Id: Ie2925ebd697e1ffa7c4cbdc9a946531f11f4d934 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39348 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Reviewed-by:
-
Ben Clayton authored
These will be used to create a synthetic file for all the commands in the command buffer, which can be stepped over using the new debugger. Bug: b/145351270 Change-Id: If5728463c1cf333d635c46e2d3b6314ce7d27326 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38910Tested-by:
-
Ben Clayton authored
Also add SPIRV-Tools, as this was missing and caused red squiggles. Bug: b/145351270 Change-Id: I22e664bb662f75284cf93cf6e446e775521c7b26 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38908Tested-by:
-
SwiftShader Regression Bot authored
Change-Id: I4c2ca30b1cc9edda07c83056d9798fe8c202152b Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39313Reviewed-by:
-
Nicolas Capens authored
The Subzero JIT now has sufficient functionality to support running the Vulkan unit tests. Bug: b/130461828 Change-Id: I8c398b976899151ac4a82549df21e06127a115f1 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39328Tested-by:
-
- 10 Dec, 2019 15 commits
-
-
Ben Clayton authored
Both code paths will be tested for Windows, macOS, Linux for both LLVM and Subzero backends. Bug: b/145836207 Change-Id: I707b1551fe1a65549c47a6d7756ab94aeb8d80e2 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39312 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
Chris Forbes <chrisforbes@google.com>
-
Chris Forbes authored
VkPhysicalDeviceFeatures::imageCubeArray implies both the ImageCubeArray and SampledCubeArray SPIRV capabilities are supported. We had only accepted SampledCubeArray so far. Bug: b/144988970 Change-Id: I93fb93a95047f73c677de38757660ea1321a4654 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39289Tested-by:
-
Ben Clayton authored
EventListener is an interface that is used to listen for thread events. Thread holds the state for a single thread of execution. Frame holds a number of variable scopes for one of a thread's stack frame. Scope is a container for frame variables. Bug: b/145351270 Change-Id: Ic61e17f32cfd6929dbd7b0fce1ffb716301fc73e Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38897Reviewed-by:
-
Ben Clayton authored
It holds a file path and line number. Bug: b/145351270 Change-Id: I0f54d00bda151be529d912562061052abff2fe81 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38896Reviewed-by:
-
Ben Clayton authored
Files represent either physical, filesystem-backed files, or virtual debugger-produced files. Bug: b/145351270 Change-Id: Ib085a1244cc10dc644546d1d049d1bfa030744ff Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38895Reviewed-by:
-
Ben Clayton authored
WeakMap is an associative container of keys of type K to values of type std::weak_ptr<V>. WeakMap's iterators will skip over elements where the value has no more remaining std::shared_ptr<V> references. Used to automatically track live debug objects. Bug: b/145351270 Change-Id: I87b9a9be4b83ff887510f292a4e03770ccc9eeaf Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38894Reviewed-by:
-
Ben Clayton authored
The Value interface exposes values to the debugger. Variables are named values. VariableContainer is a collection of Variables. Bug: b/145351270 Change-Id: Iae3749447554977f84bfc6e58f37b52aa6ed262f Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38893Reviewed-by:
-
Ben Clayton authored
This implements a very basic runtime type information used by debug-exposed values. Bug: b/145351270 Change-Id: I6bd9fe292bd7058307789300d562f7699e12f02b Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38892Reviewed-by:
-
Ben Clayton authored
set() does not evaluate conditional expressions like if() does. BUILD_CPPDAP was being set to "${BUILD_VULKAN};AND;${ENABLE_VULKAN_DEBUGGER}", so cppdap was always required as a submodule dependency. Bug: b/145351270 Change-Id: I86c3e43b8e22aa283bd2a5a20e0c03c01a965e8a Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39310Reviewed-by: -
SwiftShader Regression Bot authored
Change-Id: Ia1fb2619484845e1d7f26174baed791500493890 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39208 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Chris Forbes authored
This is required by vkmark and various other workloads. The present mode handling is not really correct -- both our fifo and mailbox modes behave more like IMMEDIATE due to XPutImage etc being completely unthrottled Future changes will introduce proper present mode handling for the Xlib and Xcb WSI implementations. Bug: b/124265819 Change-Id: Ia71ff928848be9925ffb30a27fe6c35732b68590 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39288 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Ben Clayton authored
Limit CI results to the owner when things are good. Limit CI results to owner + reviewers when things are not so good. Should reduce email spam. Change-Id: I1b71f17f306bf04ecb9fb87ffd95aea42a8e84b4 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39308Reviewed-by:
-
Antonio Maiorano authored
Made this change in code review without proper testing. The "level" being set here was the input argument, rather than the data member. Bug: b/145685985 Change-Id: I8641701cc0600837f4dc612d29ffaa4cd63390b1 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39269 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Nicolas Capens authored
CMake (without or without Visual Studio 2019) should be used instead for the Vulkan build. Bug: b/139189696 Change-Id: I8583ecf4605c89edffefa10050f7d6ef85ecb72c Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39228 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Ben Clayton authored
vk::dbg::ID is a strongly-typed identifier backed by an int. ID is used to prevent implicit casts between idenfitifers of different T types. Bug: b/145351270 Change-Id: Ia51ee8cca19eaa86cab4e009340fdcf9072360e5 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38891Reviewed-by:
-
- 09 Dec, 2019 10 commits
-
-
Chris Forbes authored
In LLVM3, this conversion was subtly broken. We're now on LLVM7, where this should be fixed. Mostly a revert of https://swiftshader-review.googlesource.com/c/SwiftShader/+/7354, but also includes a subzero implementation Bug: b/31816482 Change-Id: I154a221b44b985a35bcc5e15923a0deb9b51c711 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39148Tested-by:
-
Nicolas Capens authored
Conversion from floating-point to unsigned integer is not a native instruction on some CPU architectures, so excercise the case where it overflows signed int. Bug: b/31816482 Change-Id: I4ce924bbf4eb4c371e25d800fd4d7794edfa62cb Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39188 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Reviewed-by:
-
Antonio Maiorano authored
Allows us to configure Reactor backend optimization level via preprocessor macro. Can be set via CMake arg. Bug: b/145685985 Change-Id: I27a80b15846a8b43be5bff0505632213852a5dc6 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39168Reviewed-by:
-
Antonio Maiorano authored
Only breaks in Om1 optimization mode because this code path is specific to Om1. Broke after 01386d17 was merged because maxOutArgsSizeBytes() now returns 32 bytes for the shadow store, so this is a rarely exercised code path. Problem is that on x64, the new register being created needs to match the size of the destination register, so 64 bits, not 32 as it was assuming. Fixes ReactorUnitTests Call_ArgsMixed, Call_ArgsPointer, IntrinsicTest_* which were asstering with: Assertion failed: !isScalarIntegerType(Dest->getType()) || (typeWidthInBytes(Dest->getType()) <= typeWidthInBytes(Source->getType())), file C:\src\SwiftShader2\third_party\subzero\src\IceInstX86Base.h, line 1308 Bug: b/145685985 Change-Id: I298edd5b4a286a242b351cc6267d70ecb37943fb Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39169 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Antonio Maiorano authored
Reverted the hack fix from bb222a10, and implemented a proper fix. The fix in TargetX86Base<TraitsType>::lowerCast (IceTargetLoweringX86BaseImpl.h) was suggested by Jim Stichnoth. The original problem is that we're trying movd an i32 to an i4i32, and since the target Variable is long-lived, it may not get a register allocated for it, and when that happens, we end up tripping the assert in InstX86Movd::emitIAS() that expects the destination to have been allocated a register. The solution that Jim suggested, and is implemented here, is to create a temporary, short-lived, variable to first movd into, which should guarantee a register target, since short-lived Variables usually get registers. Then we 'mov' the temporary register Variable to Dest, which *should* support moving the i4i32 register operand to an i32 memory operand. I said *should* above, because with the above fix, we now trip another assert in InstX86Mov::emitIAS (Mov, not Movd). The reason being that it doesn't actually support moving an i4i32 reg -> i32 memory operand. I added support for this as well (IceInstX86BaseImpl.h). Note that this assert only tripped when building with Om1 optimization level, since O2 ostensibly optimized out the mov call. Bug: b/145529686 Change-Id: I3c998a3e308838123cb415fcbf9f277113ac7d28 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39068 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Ben Clayton authored
`ENABLE_VULKAN_DEBUGGER` is the new option to enable building of the Vulkan debugger, which defaults to off. `cppdap` is a new library that implements the debugger protocol. Added a stub .cpp file to try and detect any build systems that are recursively globbing into the new src/Vulkan/Debug directory. Actual debugger implementation will come in future changes. Bug: b/145351270 Change-Id: I48619f1d8352c0525c8adbaa4bc3f2373a237d61 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38890 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Ben Clayton authored
Used by the new Vulkan debugger. Bug: b/145351270 Change-Id: I4419fa208b52ed2db17422e3c14fe01dfb55cde9 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38889 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Ben Clayton authored
Required by cppdap, which is used by the new Vulkan debugger. Bug: b/145351270 Change-Id: Ibf65a63ed3c508722993578c57d72b1884a06ff9 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38888Reviewed-by:
-
Nicolas Capens authored
Reactor's 'Int' type is specified to be 32-bit, always, and it was implicitly assumed the C++ 'int' type has the same size. Likewise for 'short' being 16-bit. Make this assumption explicit with static asserts. Bug: b/145913083 Change-Id: Ie47346358487f8152f68654b75ea44b5d4f8823b Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/39129 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Reviewed-by:
-
Alexis Hetu authored
This cl adds logic to propagate the 4th texture coordinate, which is only used for cubemap arrays, and use it as the cubemap array id. Tests: dEQP-VK.glsl.texture_functions*cubearray* Bug: b/144988970 Change-Id: I241b9a50a87424461647f00d442774ead8aaa0f7 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38876 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
