- 12 Nov, 2019 6 commits
-
-
Ben Clayton authored
Use go 1.13's new error functionality to create nested errors, and use the new errors.As() to unwrap them. Fixes bad classification of tests that have recently started returning non-zero exit codes. Change-Id: I4f356696d2a7ce1576be1c350fc14e156f63c659 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38130Reviewed-by:
Chris Forbes <chrisforbes@google.com> Reviewed-by:
Nicolas Capens <nicolascapens@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
Ben Clayton <bclayton@google.com>
-
Tobin Ehlis authored
Update the maxPerStageDescriptorStorageBuffers from 4 to 16. ANGLE requires at least 12 for GLES3.1 support, and 16 is a common number among actual hardware so bumping to 16. Bug: swiftshader:136 Change-Id: I2790a8cab197c773aa14d9c1b0fdcc5ca8356b25 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38068 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
Tobin Ehlis <tobine@google.com>
-
Nicolas Capens authored
This reverts commit 5d0faf9b. Reason for revert: Marl issue was fixed; https://github.com/google/marl/issues/38 Bug: b/140575360 Change-Id: I04ca4ced82cbe0e16179cc940adc027c9a17c29f Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38048Reviewed-by:
-
Ben Clayton authored
It seems that once a change is no longer in the top N most recent changes, the server will start responding `server does not allow request for unadvertised object` when attempting to fetch the specific change by SHA. To work around this, we need to fetch the branch by name, then checkout the specific change locally. As the Khronos seems to be particularly slow to respond today, I've limited the branch depth to 99, for now. This hopefully will not be a problem as we should typically only be updating to newer changes. Also bumped the git command timeout to 15 mintutes as this repo can be _really_ slow. Change-Id: Ie08165a6ca92e79d56bb4d2d5376410f2850a471 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38129Tested-by:
-
SwiftShader Regression Bot authored
Change-Id: I9c4f1e6716ff12d05de18479703b004370b93bbd Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38128 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Alexis Hetu authored
For 565 and 5551 formats, the blend equations were getting applied on top of colors represented by the top 5 or 6 MSB only, leading to precision errors. Fixed this by: - removing the scaling applied on the source color before blending is applied and the related inverse transform in the writeColor function - adding the proper remapping of the color to the full color range in the readPixel function That way, we're always working with full 16 bit range colors and the blend equations work properly. Fixes all tests in: dEQP-GLES3.functional.fragment_ops.interaction.basic_shader.* Bug: angleproject:4016 b/24332884 Change-Id: I53531b88a8629aaa61929e99efaac38cdb46c834 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38088Tested-by:
Alexis Hétu <sugoi@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 11 Nov, 2019 6 commits
-
-
Tim Van Patten authored
This change allows for outputting the SwiftShader version to Android's logcat. The version information is as follows: major.minor.patch.commit_hash In logcat, when SwiftShader is loaded by ANGLE, this looks like: 11-07 12:17:10.152 9110 9110 I SwiftShader: SwiftShader Version: 5.0.0.9c7bf8093a75 11-07 12:17:10.162 9110 9110 I ANGLE : Version (2.1.0.a2b1a958e2d8), Renderer (Vulkan 1.1.0(SwiftShader Device (0x0000C0DE))) This allows developers to verify that they are running the expecting build of SwiftShader when debugging problems or running conformance tests. The version information is logged to logcat when the build flag "ENABLE_BUILD_VERSION_OUTPUT" is enabled within src/Android.bp. Bug: b/142828252 Change-Id: Iff773b16a2f3532aa843629ec50519b519bbadac Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37990 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
Tim Van Patten <timvp@google.com> Reviewed-by:
-
Chris Forbes authored
Test: dEQP-VK.renderpass.suballocation.unused_clear_attachments.*depth* Bug: b/144280552 Change-Id: I37072270105caf206ed3ab6cb2895d32ef22f653 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38108Tested-by:
-
SwiftShader Regression Bot authored
Change-Id: I9923ef71a77ee27fa8cb276e3aa924ca23ce08fd Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37968Reviewed-by:
-
Chris Forbes authored
The module may legitimately contain descriptor set references (intended for another entrypoint) which are not compatible with our limits. There is a matching assert to verify the set index at the point the descriptor is actually referenced. Bug: b/140648941 Change-Id: Ie3b8c6a2e53e553d1dccb2995e6719539eab89de Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36114Tested-by:
-
Alexis Hetu authored
Moved some functionality from libVulkan.cpp to VkImage.cpp in order to be able to remove the global androidSwapchainMap which may have been causing sync issues. I tried to make as much code as possible compile on desktop so that we have a better chance of detecting if a desktop related change breaks Android specific code. Bug: b/143902189 Change-Id: Ie9e17a77c2a1705b3194ebd67adeb1a3c0e6213f Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37828 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Reviewed-by:
-
Ben Clayton authored
Fixes the loading of the dEQP config file from the CI change. Bug: b/140172528 Change-Id: Iabe0b66fd305cdd9c8bb0869c1ea8082cf7d236c Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38029 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 09 Nov, 2019 1 commit
-
-
Chris Forbes authored
(a7d268a6a3ee9655298d61cdb8b5c42306a4e335) Bug: b/144183516 Change-Id: Ib4497dc3aaa033904aa269a33244e2eb3b5b6cf1 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38069Reviewed-by:
-
- 08 Nov, 2019 2 commits
-
-
Chris Forbes authored
There were two bugs here: - Across a 2x2 pixel quad, we were producing a uniform PointCoord value. This is not correct, and causes trouble even for size 1 points if the derivative of the PointCoord value is ever used, or if the non-helper lane is not lane 0 (in which case the live lane may end up having PointCoord values outside [0,1].) - SetupRoutine was using the clipspace position of the vertex. Xf,Yf are specified to be in screenspace. It's possible to do this in a less redundant way, but currently we don't have unrounded vertex positions in screenspace available in the setup program. Just reproject for now, as we do in clipping paths etc. Note that we only do this if we're drawing points, so most (filled triangle) draws should not incur any additional cost. Bug: b/140680704 Test: dEQP-VK.glsl.builtin_var.simple.pointcoord* Change-Id: If006cc92b86673b6bb747cf60f60e0ed46f5e8e3 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/36115Tested-by:
-
Ben Clayton authored
This adds to 'Bug:', 'Issue:' and 'Regres:' as accepted CL grep checks. Fixes: b/143093187 Change-Id: Iae44d8e8e49f7534bb940ec977edb4afcddc308e Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37449Reviewed-by:
-
- 07 Nov, 2019 1 commit
-
-
Alexis Hetu authored
The scissor is now restricted to the framebuffer's size, in order to avoid writing out of bounds of the image. Vulkan 1.1 spec section 25.2. Scissor Test states that "Rasterization does not produce fragments outside of the framebuffer" Bug: b/141999739, angleproject:4060 Change-Id: Id0bce035b0ab9072f354b9f00be8d42577e4af54 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/38008Reviewed-by:
-
- 06 Nov, 2019 2 commits
-
-
Jonah Ryan-Davis authored
MetalSurfaceEXT and MacOSSurfaceMVK were inheriting incorrectly from the base class MetalSurface, and were improperly initialized. Additionally, none of the drawables were ever released, meaning there were no re-used drawables. Add an @autoreleasepool around the drawable retrival path, as per the spec: https://developer.apple.com/documentation/quartzcore/cametallayer?language=objc#3172024 Bug: chromium:1015454 Change-Id: Idd2f1842977a6f5dccecf31d152d454a44e9261c Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37948Reviewed-by:
Jonah Ryan-Davis <jonahr@google.com>
-
SwiftShader Regression Bot authored
Change-Id: I81aa8a7fdfb1c900062cc68f030e8765e9f019a2 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37528Reviewed-by:
-
- 05 Nov, 2019 2 commits
-
-
Jonah Ryan-Davis authored
VK_EXT_metal_surface is a similar extension to VK_MVK_macos_surface, which is already supported. Bug: chromium:1015454 Change-Id: I08d8e6b097d75236f724984b8c7d20251a63581f Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37808 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Nicolas Capens authored
Copied from android/cts/master/vk-master.txt at dEQP revision 76a2b81e0ca86a65a927892ee4340bc6685e86b3 from gitlab.khronos.org/Tracker/vk-gl-cts.git, which is the revision currently used by Regres (see tests/regres/deqp.json). WSI tests were removed from the Android vk-master.txt, as they only contain Android WSI tests. Instead Regres uses vk-wsi.txt, which doesn't need updating. Bug: b/140172528 Change-Id: I2b6ea4cdf4626338740e32fe2792998dd21eb157 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37888 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Reviewed-by:
-
- 01 Nov, 2019 8 commits
-
-
Antonio Maiorano authored
Bug: b/143479561 Change-Id: I25e68cc24f6141611493ea7dc24f90ab4b2e4d4b Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37711Tested-by:
Antonio Maiorano <amaiorano@google.com> Reviewed-by:
-
Antonio Maiorano authored
Bug: b/143479561 Change-Id: I555db2de99dfc5a139fbdf61c8ac4447e58ffd02 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37709Reviewed-by:
-
Antonio Maiorano authored
Bug: b/143479561 Change-Id: Iedd6b62b5b3dcc46a789f15112656cfe191cb4de Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37708Tested-by:
-
Antonio Maiorano authored
Bug: b/143479561 Change-Id: I1a63f7eab9d5b6ce12f5d858018c4c6582ab9c90 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37669 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Antonio Maiorano authored
Bug: b/143479561 Change-Id: Id6039b31790eb181daf669dff43cd46f92d464d7 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37748Tested-by:
-
Antonio Maiorano authored
Bug: b/143479561 Change-Id: Ie1b8dad6611d5fb1e2a6e4cc679c01ac98cf8e4d Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37668 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Jason Macnak authored
... by updating eglSwapBuffers() to give up the display lock before calling surface->swap(). Bug: b/135742693 Test: built locally Test: presubmit Test: cts -m CtsMediaTestCases -t android.media.cts.HeifWriterTest#testInputSurface_Grid_Handler Test: dEQP-GLES3 tests through cherry Change-Id: I816d30e80077b40e8396bf8d6ae5bb236753e3ea Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37769Reviewed-by:
Jason Macnak <natsu@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
-
Jason Macnak authored
The next change will update the locking around eglSwapBuffers. Move all the locking into libEGL.cpp to avoid duplicating getDisplayLock(). Bug: b/135742693 Test: built locally Test: presubmit Test: cts -m CtsMediaTestCases -t android.media.cts.HeifWriterTest#testInputSurface_Grid_Handler Change-Id: I1ca863b4902d1dc4e456700ac96aecc3e61424ba Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37768 Presubmit-Ready: Jason Macnak <natsu@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 30 Oct, 2019 2 commits
-
-
David 'Digit' Turner authored
This corresponds to the VK_FUCHSIA_external_semaphore extension which uses a Zircon event object to communicate across processes. Bug: b/140421726 Change-Id: I47b235d4ff7d787491738422bda6fdf853803ab7 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/35969 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
David Turner <digit@google.com>
-
David 'Digit' Turner authored
A previous CL updated the Vulkan headers, and removed our include for vulkan_fuchsia_extras.h, which contains Fuchsia-specific declarations that are used by the platform but not upstreamed yet to Khronos. This CL simply restores the include. Bug: 143197472 Change-Id: I934f185c76f54e08ff52a202198cb951ad4b6644 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37450Tested-by:
-
- 29 Oct, 2019 2 commits
-
-
Alexis Hetu authored
A mistake found its way into the previous "Provoking vertex fixes" cl. This cl fixes the vkGetPhysicalDeviceFeatures2 function for the VK_EXT_provoking_vertex extension. Bug: angleproject:3677, angleproject:3430 Change-Id: I561361f3fd2c3f6b7ca524614b7d7af461cc12d4 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37710 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Reviewed-by:
-
Alexis Hetu authored
A few things were buggy in the last provoking vertex cl: - There were 2 separate loops for rasterizationState extensions. Correctly combined them into a single loop. - There was a missing break statement in vkCreateDevice - There was a missing case in vkGetPhysicalDeviceFeatures2 Bug: angleproject:3677, angleproject:3430 Change-Id: Id4a16168933fc8c440b7339851a39e8cc4683491 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37688 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Reviewed-by:
-
- 28 Oct, 2019 2 commits
-
-
Alexis Hetu authored
Adding an new extension, which is not present in the Vulkan headers, in order for ANGLE to be able to specify to SwiftShader which provoking vertex convention to use, in order to properly emulate OpenGL ES on top of Vulkan, since OpenGL ES uses the last vertex convention and Vulkan uses the first vertex convention. Bug: angleproject:3677, angleproject:3430 Change-Id: If4e6539d48bc282af8ed6f2a9db6b2fc9d0dbe2d Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37248 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Jonah Ryan-Davis authored
MacOS requires a few defines that were missing from BUILD.gn. It also requires being linked to Cocoa/IOSurface/QuartzCore. It also needs macros to define API availability to use CAMetalLayer. Bug: chromium:1015454 Change-Id: I6b298cc805ce1f927d2eed8f4d0e11a271f75f24 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37628 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 25 Oct, 2019 2 commits
-
-
Alexis Hetu authored
Changing the stencil reference shouldn't be causing a new fragment shader to compile, since it has no effect on the generated program. This cl removes the front and back stencil reference from the States structure to avoid uselessly generating programs. Improves the run times of a few dEQP.KHR_GLES3 tests with SwANGLE: dEQP.KHR_GLES3/packed_depth_stencil_verify_read_pixels_depth24_stencil8 -> Test runs ~8X faster (from ~4s down to ~500ms) dEQP.KHR_GLES3/packed_depth_stencil_verify_read_pixels_depth32f_stencil8 -> Test runs ~19X faster (from ~3.8s down to ~200ms) Bug: b/143288278 Change-Id: Ic46db6822daa924599de0f0a9e9688f8bacc95ea Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37568 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Antonio Maiorano authored
Manually applied https://reviews.llvm.org/rL357714 to fix crash/assert on loading SwiftShader when compiled with VS 2019 16.3.29324 and later, and using clang-cl to compile. More information about this bug: https://bugs.llvm.org/show_bug.cgi?id=41367 Bug: chromium:1015904 Change-Id: I83ad7feb15a6ca5d5eb0fc25fb6674d4678d60a5 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37608 Presubmit-Ready: Antonio Maiorano <amaiorano@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 24 Oct, 2019 3 commits
-
-
Alexis Hetu authored
The destination offset wasn't taken into account in the computation of the source offset, which is necessary because we compute the source coordinates from the destination coordinates in the generated routine. Bug: angleproject:4037 angleproject:4013 Change-Id: Ib9961cec267653c8d33da1847486943f8a670a49 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37588 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Tested-by:
-
Alexis Hetu authored
True integer types don't support MSAA, so the blitter never had support for MSAA on true integer types. In Vulkan, VK_FORMAT_S8_UINT is a UINT format, so it's considered a true integer type, but it must still support MSAA. Added support for MSAA on true integer types in the blitter so fix all related tests with SwANGLE: dEQP-GLES3.functional.depth_stencil_clear.* dEQP-GLES3.functional.fragment_ops.* dEQP-GLES3.functional.rasterizer_discard.* Bug: angleproject:4044 Change-Id: I8e04f434f0f671999aaef23cf24cc6684fd0fc16 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37548 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Paul Thomson authored
...to reduce file size of debug builds on Linux. Bug: b/142862126 Change-Id: Idab23a044502cfebb57e6b912a6c1f95692c50c8 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37388 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
Paul Thomson <paulthomson@google.com>
-
- 23 Oct, 2019 1 commit
-
-
Sascha Willems authored
This adds an ICD entrypoint for vkGetPhysicalDeviceWin32PresentationSupportKHR to the Win32 platform. This change will allow tools that make use of this function to work with the swiftshader ICD. One such tool is my Vulkan Hardware Capability viewer, that would crash on ICDs not implementing this. Bug: b/143207519 Change-Id: I23552cf072595b97f3032416a51c0db13d3f87a9 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/37509 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
