- 06 Jul, 2020 1 commit
-
-
Nicolas Capens authored
VK_FORMAT_B10G11R11_UFLOAT_PACK32 is a mandatory uniform texel buffer format, which we claimed support for, but the implementation was missing. Bug: b/159561781 Change-Id: I47a1aa3d4f371603e4d97e85526e10677753b9ca Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/46008 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Tested-by:
Nicolas Capens <nicolascapens@google.com> Reviewed-by:
Antonio Maiorano <amaiorano@google.com>
-
- 02 Jul, 2020 2 commits
-
-
Alexis Hetu authored
A few struct members were only ever written to and never read, which means these were encoder only struct members. Since we only need the ASTC decoder, these encoder only struct members were removed. The gains in memory allocation sizes are substantial: block_size_descriptor Initial size: 3469056 B (~3.3 MB) After this cl: 693168 B (~677 kB) decimation_table Initial size: 364896 B (~356 kB) After this cl: 1948 B (~1.9 kB) Change-Id: Id86c993331d2602a1106f6aaa729d1047e2ebc27 Tests: dEQP-VK.*astc* Bug: b/150130101 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/46128 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Reviewed-by:
Alexis Hétu <sugoi@google.com>
-
Nicolas Capens authored
It was being set as a C++ compile flag, while it's an MSVC linker flag. The /D was interpreted as a macro definition, which caused an IntelliSense error. Also fix the set_linker_flag CMake macro to correctly append the build configuration. Bug: b/147735529 Change-Id: I0e4473a07c224b64ad0aeb54390bced087eea3d8 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/46088Reviewed-by:
-
- 01 Jul, 2020 1 commit
-
-
Peter Collingbourne authored
This is a cherry pick of upstream LLVM r340889. Bug: b/74603866 Change-Id: I9d0fd16d20d690b5c930e0a915dad8247890fc38 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/46048Tested-by:
Peter Collingbourne <pcc@google.com> Presubmit-Ready: Peter Collingbourne <pcc@google.com> Kokoro-Result: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 30 Jun, 2020 1 commit
-
-
Sean Risser authored
The original fix for b/139528538 didn't differentiate between release and debug builds properly. It just just checked to see if a debugger was attached. Now the assert will only happen when a debugger is attached and in a debug build. Bug: b/159733084 Change-Id: I9a6407685225c662aeb378be3296b8e4640b279d Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/46028 Kokoro-Result: kokoro <noreply+kokoro@google.com> Tested-by:
Sean Risser <srisser@google.com> Reviewed-by:
-
- 28 Jun, 2020 1 commit
-
-
Nicolas Capens authored
Extend Blitter::fastClear(), which is based on memset() instead of using Reactor routines, to also handle D32_SFLOAT and S8_UINT formats. Benchmark results: Run on (48 X 2594 MHz CPU s) CPU Caches: L1 Data 32 KiB (x24) L1 Instruction 32 KiB (x24) L2 Unified 256 KiB (x24) L3 Unified 30720 KiB (x2) -------------------------------------------------------------------------- Benchmark Time CPU Iterations -------------------------------------------------------------------------- (LLVM, before) ClearImage/VK_FORMAT_D32_SFLOAT 3.74 ms 0.016 ms 1000 (LLVM, after) ClearImage/VK_FORMAT_D32_SFLOAT 1.08 ms 0.044 ms 10000 (Subzero, before) ClearImage/VK_FORMAT_D32_SFLOAT 4.51 ms 0.063 ms 1000 (Subzero, after) ClearImage/VK_FORMAT_D32_SFLOAT 0.963 ms 0.040 ms 7467 This change re-implements https://swiftshader-review.googlesource.com/c/SwiftShader/+/45888 which was reading 'clearValue' out of bounds when accessing color[1] when it's only a single depth or stencil value. Bug: b/159455503 Bug: chromium:1097740 Change-Id: Id3e74b4fa28ee0422540a8480814f8c9988f402a Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45949 Kokoro-Result: kokoro <noreply+kokoro@google.com> Tested-by:
-
- 26 Jun, 2020 1 commit
-
-
Nico Weber authored
Bug: chromium:1098899 Change-Id: Iad5cd599e7e90f29249d0e4129ca1fe4a10d77c6 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/46029 Presubmit-Ready: Nico Weber <thakis@google.com> Kokoro-Result: kokoro <noreply+kokoro@google.com> Reviewed-by:
Nico Weber <thakis@google.com>
-
- 23 Jun, 2020 3 commits
-
-
Antonio Maiorano authored
This will help ensure we don't break these configs. Bug: b/159633249 Change-Id: Ida7074a649c10a1a5b35ce82b4f8438b65bdb3eb Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45970 Kokoro-Result: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Antonio Maiorano authored
Also simplify code so that emitPrintLocation is called, but does nothing if ENABLE_RR_EMIT_PRINT_LOCATION isn't defined. Bug: b/159633249 Change-Id: I0e207baf6a0e2eb144e026904357664e1277ebf1 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45969 Kokoro-Result: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Antonio Maiorano authored
This is as close as we'll get to "set -e" from bash. Makes the script less verbose, and closer to the Linux and MacOS ones. Bug: b/159633249 Change-Id: Ied7cdb5ab11b737fb21379be912393bbd838a40a Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45968 Kokoro-Result: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 22 Jun, 2020 1 commit
-
-
Corentin Wallez authored
This reverts commit af1f2159. Reason for revert: Fails tests with ASAN crbug.com/1097740 Bug: chromium:1097740 Change-Id: I48d3d1129ad3ea314df29d1f40b595ad6fa169ce Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45928Reviewed-by:
Corentin Wallez <cwallez@google.com> Tested-by:
-
- 20 Jun, 2020 2 commits
-
-
Nicolas Capens authored
The Vulkan spec states that "blendEnable [...] must be VK_FALSE if the attached image’s format features does not contain VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT" Previously we would silently ignore blending of integer formats. It was determined to be undefined behavior in https://gitlab.khronos.org/vulkan/vulkan/issues/2098 Bug: b/155147929 Change-Id: I01500d8c39d7f2c9a484944b4a93c6004e938c05 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45708 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Kokoro-Result: kokoro <noreply+kokoro@google.com> Tested-by:
-
Nicolas Capens authored
A highly generic Blitter::blitToBuffer() method was used for copying pixels without requiring any format conversion or scaling. This change removes Blitter::blitToBuffer() and the unused blitFromBuffer() methods and implements a copyTo() method which does straightforward memcpy() of the data. Also consistently add 'To' and 'From' to the Image class's copy, blit, and resolve methods to avoid any confusion about which direction the transfer happens. Benchmark results: Run on (48 X 2594 MHz CPU s) CPU Caches: L1 Data 32 KiB (x24) L1 Instruction 32 KiB (x24) L2 Unified 256 KiB (x24) L3 Unified 30720 KiB (x2) --------------------------------------------------------- Benchmark Time CPU Iterations --------------------------------------------------------- (LLVM, before) Triangle/Hello 8.69 ms 7.39 ms 112 (LLVM, after) Triangle/Hello 0.878 ms 0.454 ms 2133 (Subzero, before) Triangle/Hello 26.2 ms 24.8 ms 41 (Subzero, after) Triangle/Hello 1.11 ms 0.432 ms 1339 Bug: b/147967959 Change-Id: I76a103bbb6e582f987ef1c3cc39d17779ae4ac99 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45650 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Kokoro-Result: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 19 Jun, 2020 1 commit
-
-
Nicolas Capens authored
Extend Blitter::fastClear(), which is based on memset() instead of using Reactor routines, to also handle D32_SFLOAT and S8_UINT formats. Benchmark results: Run on (48 X 2594 MHz CPU s) CPU Caches: L1 Data 32 KiB (x24) L1 Instruction 32 KiB (x24) L2 Unified 256 KiB (x24) L3 Unified 30720 KiB (x2) -------------------------------------------------------------------------- Benchmark Time CPU Iterations -------------------------------------------------------------------------- (LLVM, before) ClearImage/VK_FORMAT_D32_SFLOAT 3.74 ms 0.016 ms 1000 (LLVM, after) ClearImage/VK_FORMAT_D32_SFLOAT 1.08 ms 0.044 ms 10000 (Subzero, before) ClearImage/VK_FORMAT_D32_SFLOAT 4.51 ms 0.063 ms 1000 (Subzero, after) ClearImage/VK_FORMAT_D32_SFLOAT 0.963 ms 0.040 ms 7467 Bug: b/159455503 Change-Id: Ib1b0ad38417c516267d6addd835076a933e22dba Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45888 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Kokoro-Result: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 18 Jun, 2020 3 commits
-
-
Nicolas Capens authored
The spec for VkDescriptorSetLayoutBinding states that "If descriptorCount is zero this binding entry is reserved and the resource must not be accessed from any stage via this binding within any pipeline using the set layout." This restores the debug validation that existed at https://swiftshader-review.googlesource.com/c/SwiftShader/+/44308/4/src/Pipeline/SpirvShaderMemory.cpp#b405 Bug: b/157915024 Change-Id: I5e6798b0cc87f564a3c6f13963995822660b5111 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45468Tested-by:
-
Nicolas Capens authored
Also separate the out-of-bounds handling from the offset calculations. Bug: b/155862459 Change-Id: Ib6c5ea029d16fc411d792e6c129b37386d48fc1d Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45092Tested-by:
-
Nicolas Capens authored
Adding large values can overflow the already out-of-bounds offsets to become negative. Negative offsets are not detected as out-of-bounds, so it would result in an out-of-bounds access. This can also happen when they were already large negative values and remain negative after the addition. Or, when they're small negative values, they can become in bounds and thus not trigger the desired out-of-bounds behavior. Just replace the offsets with the out-of-bounds values. Bug: b/155862459 Change-Id: I38a45832bfea3a13a66f133ee7c2162855f02c61 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45091 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Kokoro-Result: kokoro <noreply+kokoro@google.com> Tested-by:
-
- 17 Jun, 2020 6 commits
-
-
Nicolas Capens authored
This refactors the benchmarks to use static functions which support passing arbitrary arguments, instead of fixtures: https://github.com/google/benchmark#passing-arbitrary-arguments-to-a-benchmark This makes it easier to have a variant with and without multisampling, and avoids using state.counters[] as a hack to display clear formats. Bug: b/158231104 Change-Id: Icf0eed254dc942af8f2bcfc975a1e5e581b6e786 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45790 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Tested-by:
-
Nicolas Capens authored
Adds a Framebuffer class to encapsulate and abstract framebuffer creation with or without multisampling. Bug: b/158231104 Change-Id: I4198f53f872aae76955ba13f1f2372067b91a07c Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45789 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Tested-by:
-
Nicolas Capens authored
For now, this has to be enabled manually using TriangleBenchmark::multisample Bug: b/158231104 Change-Id: I929da4bf72e6f354aeb089e19fdd6a749f839d3f Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45689 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Tested-by:
-
Nicolas Capens authored
Initialization of thread_local variables in shared libraries might be implementation dependent. Just keep the assert for Win32. https://docs.microsoft.com/en-us/cpp/cpp/thread?view=vs-2019 states that "Thread-local variables that are initialized statically with constant values are generally initialized properly on all threads." Bug: b/157525646 Change-Id: Ifb6a77e085c0f21f13af58c89050868992928300 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45809 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Tested-by:
-
Nicolas Capens authored
The Vulkan specification for VkGraphicsPipelineCreateInfo states: * pViewportState is a pointer to a VkPipelineViewportStateCreateInfo structure, and is ignored if the pipeline has rasterization disabled. * pMultisampleState is a pointer to a VkPipelineMultisampleStateCreateInfo structure, and is ignored if the pipeline has rasterization disabled. * pDepthStencilState is a pointer to a VkPipelineDepthStencilStateCreateInfo structure, and is ignored if the pipeline has rasterization disabled or if the subpass of the render pass the pipeline is created against does not use a depth/stencil attachment. * pColorBlendState is a pointer to a VkPipelineColorBlendStateCreateInfo structure, and is ignored if the pipeline has rasterization disabled or if the subpass of the render pass the pipeline is created against does not use any color attachments. Also remove the UNSUPPORTED() for pTessellationState since it is valid for the application to provide a non-null pointer even though tessellation is not supported. Bug: b/155359026 Change-Id: Ic1cd9224b159cca66f0dbf1b1a563efc56386866 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45728 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Tested-by:
-
Nicolas Capens authored
Using an initialization method or in-class initialization of each member is bug-prone. Zero-initialization performed by the default constructor guarantees repeatability. Also don't skip drawing when the multisample mask is zero. Pipeline stages prior to fragment processing may have side-effects. Bug: b/155359026 Change-Id: Ibab131a41282fbdc4f5f84208776725eb3c786bd Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45748 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Tested-by:
-
- 16 Jun, 2020 8 commits
-
-
Alexis Hetu authored
VkImage use different structures to represent subregions: - VkImageSubresource (Single mip level, single layer) - VkImageSubresourceLayers (Single mip level, multiple layers) - VkImageSubresourceRange (Multiple mip levels, multiple layers) This cl changes the subregion structure used in vk::Image and in the Blitter so that it contains the information required for the current task and no more than that. Bug: b/159045555 Change-Id: I186a90ff321318319c01a0718a95bbae2d2eb26a Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45788 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Reviewed-by:
-
Ben Clayton authored
Changes: d29553a3730 Add missing <vector> include Commands: ./third_party/update-marl.sh --squash Bug: b/140546382 Change-Id: Ibb19c64cd2cbbd4b71ddd8270d1b3541a123cbca -
Ben Clayton authored
d29553a3730 Add missing <vector> include git-subtree-dir: third_party/marl git-subtree-split: d29553a3730f2bbe7ab4afd37e926e787faae4a0
-
Ben Clayton authored
Changes: 5c6c1532220 Fix Android build Commands: ./third_party/update-marl.sh --squash Bug: b/140546382 Change-Id: Ibb733d85c64a183f4fbe1a803d3d569a37e858a1 -
Ben Clayton authored
5c6c1532220 Fix Android build git-subtree-dir: third_party/marl git-subtree-split: 5c6c1532220096dd470549312053a445d7d95a2c
-
Ben Clayton authored
* changes: Update Marl to c51271125 Squashed 'third_party/marl/' changes from 38c0c7a0f..c51271125 Repoint marl branch from master to main SubzeroReactor: Use new marl::Scheduler::Config API.
-
Ben Clayton authored
In-memory data visualizations of vector types were using indices for displaying vector elements. For non-memory data, we use .x .y .z .w for vectors <= 4 elements. Mirror this for in-memory data. Bug: b/148401179 Change-Id: Ib2e8e2019ab274dfae2d6e18978f41c721c800a1 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45609 Kokoro-Result: kokoro <noreply+kokoro@google.com> Reviewed-by:
Ben Clayton <bclayton@google.com>
-
Ben Clayton authored
to debug::Object. This used to be unnecessary as the debug objects used to contain POD, but we now also have std::string and std::vector fields in these objects, and we need them to destruct. Bug: b/148401179 Change-Id: I28637f645a96e3e895e7ae5f3028eb7872c92dd3 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45608 Kokoro-Result: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 15 Jun, 2020 5 commits
-
-
Ben Clayton authored
Changes: c51271125 README.md: Repoint link from `master` to `main` da0a9c610 Add basic CMake rules for versioning marl 0e639c3c7 Benchmarks: Simplify tests further. 32af8bb50 Kokoro: Enable debug checks 8cf8dc033 Wrap all stl containers with a marl::StlAllocator 325b072b9 Support specifying worker thread affinities 49fe9a17f marl::containers: Add const methods 11f31bfbe Benchmarks: Warn if benchmarking with sanitizers bfdf613e7 Benchmarks: Add MARL_FULL_BENCHMARK flag 5e2383370 Benchmarks: Allow running with custom Config c277c61b0 marl::containers::vector fixes fcbe1f279 Add issue number to MARL_DEPRECATED() 1f010cad7 Warn about use of deprecated APIs d2e553bff Don't use deprecated scheduler methods Commands: ./third_party/update-marl.sh --squash Bug: b/140546382 Change-Id: I3e59c425cbdc868e5de44a00a5523392044a3352 -
Ben Clayton authored
c51271125 README.md: Repoint link from `master` to `main` da0a9c610 Add basic CMake rules for versioning marl 0e639c3c7 Benchmarks: Simplify tests further. 32af8bb50 Kokoro: Enable debug checks 8cf8dc033 Wrap all stl containers with a marl::StlAllocator 325b072b9 Support specifying worker thread affinities 49fe9a17f marl::containers: Add const methods 11f31bfbe Benchmarks: Warn if benchmarking with sanitizers bfdf613e7 Benchmarks: Add MARL_FULL_BENCHMARK flag 5e2383370 Benchmarks: Allow running with custom Config c277c61b0 marl::containers::vector fixes fcbe1f279 Add issue number to MARL_DEPRECATED() 1f010cad7 Warn about use of deprecated APIs d2e553bff Don't use deprecated scheduler methods git-subtree-dir: third_party/marl git-subtree-split: c51271125451c599efb9ec58b355a4c434296a8f
-
Ben Clayton authored
See https://github.com/google/marl/issues/153 for more details Bug: b/140546382 Change-Id: I73cd3e575b781890e82eda94285e8a585bd879d5
-
Ben Clayton authored
`marl::Scheduler::setWorkerThreadCount()` and `marl::Scheduler::setThreadInitializer()` have been deprecated. `marl::Scheduler::Config` is the new API. This got missed from 45493. See https://github.com/google/marl/issues/139 for context. Bug: b/140546382 Change-Id: Ic09c196b4c81f316cbf73de5395757108c16945c
-
Antonio Maiorano authored
This change now also makes it so that when rr::Print is enabled, it will output to stdout and to the debugger, if available. For now, that means it calls OutputDebugString on Windows. This is useful when it is difficult to capture stdout (e.g. debugging Chrome). Also, you can now modify Reactor.cpp and uncomment `#define RR_PRINT_OUTPUT_TYPE_STUB` to stub out the function itself. This is useful in combination with ENABLE_RR_EMIT_PRINT_LOCATION in that debugging the code-gen assembly allows you to inspect the string argument to rr::DebugPrintf and see exactly where you are in Reactor code without paying the performance cost of actually printing. Bug: b/158678105 Change-Id: Ie1bd85888405e189200c0e5f8f3449551327e0ab Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45669 Kokoro-Result: kokoro <noreply+kokoro@google.com> Tested-by:
-
- 11 Jun, 2020 4 commits
-
-
Alexis Hetu authored
The 37 failures in dEQP-VK.texture.explicit_lod.2d.* originated from a dEQP test issue which is now fixed: https://github.com/KhronosGroup/VK-GL-CTS/commit/ad89e118084d08fa1f14ad5801b996c62eb113cc Bug: b/151214291 Change-Id: I383a859945141ed04e4c49bab47a43692d02c060 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45668Reviewed-by:
-
Nicolas Capens authored
https://www.khronos.org/registry/vulkan/specs/1.2-extensions/man/html/VkShaderModuleCreateInfo.html states that the "code must adhere to the validation rules described by the Validation Rules within a Module section of the SPIR-V Environment appendix" This makes it the application's responsibility to only pass valid SPIR-V code to vkCreateShaderModule. However, improperly validated code may cause bugs which are hard to root cause, so validate in our implementation as well in debug builds. This can also help with fuzzing efforts. Bug: b/158228522 Change-Id: Ib4f41fa70a021fb902e1f8504005b87c92286942 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45569Tested-by:
-
Nicolas Capens authored
This benchmark renders a single triangle to a double-buffered swapchain. It uses a passthrough vertex and fragment shader written in GLSL and compiled into SPIR-V using glslang. Bug: b/158231104 Change-Id: I21f00f8f8595e018ff0d05f5af898a1f64d4dd12 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45629 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Tested-by:
-
Nicolas Capens authored
Note glslang's build targets may already have been added by the PowerVR SDK. In that case, we use their copy. This is not optimal since it potentially makes our test results depend on whether or not building the PowerVR samples has been enabled, but it prevents duplicate targets. Bug: b/158655757 Change-Id: Ic4e9292da9b1bb498164929cc7f93e73bf1fd6e6 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45628 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Tested-by:
-
