- 18 May, 2019 3 commits
-
-
Nicolas Capens authored
The SPIR-V spec states: "Coordinate must be a scalar or vector of floating-point type. It contains (u[, v] … [, array layer]) as needed by the definition of Sampled Image. It may be a vector larger than needed, but all unused components will appear after all used components." Bug b/129523279 Tests: dEQP-VK.glsl.texture_functions.texture.* Change-Id: I951f679b7794fef5fbbe504c753db052c5cf124b Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31694 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
Nicolas Capens <nicolascapens@google.com> Reviewed-by:
Chris Forbes <chrisforbes@google.com>
-
Nicolas Capens authored
Divide Dref by the projection coordinate. The SPIR-V spec states that "Dref /q is the depth-comparison reference value." Bug b/129523279 Tests: dEQP-VK.glsl.texture_functions.textureproj.* Change-Id: Ib1c4ad09f829b84b6b1065ff604fe88f8b1c1426 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31693 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Nicolas Capens authored
The imageView's baseMipLevel is already taken into account when obtaining each level's parameters, so don't offset a second time. Bug b/129523279 Tests: dEQP-VK.texture.mipmap.* Change-Id: I5a5c35f36d2f8517136c095f8c80202984a777bd Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31690 Presubmit-Ready: Nicolas Capens <nicolascapens@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 17 May, 2019 12 commits
-
-
Hernan Liatis authored
The swapchain images were being freed with an allocator despite never being created with one. Tests: dEQP-VK.wsi.xlib.swapchain.simulate_oom.* Bug: b/132451407 Change-Id: I26c6ecd18d04f514cdae171100de6a89c919a777 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31692 Presubmit-Ready: Ben Clayton <bclayton@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
Hernan Liatis <hliatis@google.com>
-
Hernan Liatis authored
Physical device was using the wrong allocations scope. Tests: dEQP-VK.wsi.xlib.surface.create_custom_allocator Tests: dEQP-VK.wsi.xlib.surface.create_simulate_oom Bug: b/132451407 Change-Id: I242d97548bdf01cbdab5cc708469bca02a845a66 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31691 Presubmit-Ready: Ben Clayton <bclayton@google.com> Reviewed-by:
-
Fabrice de Gans-Riberi authored
Fuchsia has no mallinfo() or getrusage(). This silences the build warnings. Bug: chromium:921807 Change-Id: Ifbb274efdc99ebdf38a6b6b8760ffd42be6fafa4 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31708Reviewed-by:
Fabrice de Gans-Riberi <fdegans@chromium.org> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com>
-
Alexis Hetu authored
Unnormalized texture coordinates are pretty much the same as rectangle texture coordinates, with a few extra addressing modes to support. ADDRESSING_CLAMP and ADDRESSING_BORDER work properly. Other modes can't used with unnormalized texture coordinates, according to the Vulkan spec: "If unnormalizedCoordinates is VK_TRUE, addressModeU and addressModeV must each be either VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE or VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER" Bug b/129523279 Change-Id: I3e39c16172bc5825ec48c6395f52e9b177df1304 Tests: dEQP-VK.texture.filtering.unnormal.* Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31648Tested-by:
Alexis Hétu <sugoi@google.com> Presubmit-Ready: Alexis Hétu <sugoi@google.com> Reviewed-by:
-
Alexis Hetu authored
If an image is created with flags including the VK_IMAGE_CREATE_BLOCK_TEXEL_VIEW_COMPATIBLE_BIT bit, then the compressed image may be used with an image view which has an uncompressed format that is size-compatible with the image's compressed format. Tests: dEQP-VK.image.texel_view_compatible.*.texture_read.e* Bug b/119620767 Change-Id: Ia1af959780af12e9014f7852aaf991d313f75bf1 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31620Tested-by:
-
Ben Clayton authored
Change-Id: Idb051ef0af8510cbb65dffbbe94be34951928c48 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31670 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
Ben Clayton <bclayton@google.com>
-
Ben Clayton authored
They're not just a Clang thing. Change-Id: Ibf66e13997dbd59ec7b1d4f9bbaf0d20ea244200 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31669 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Ben Clayton authored
Should provide more context when ASAN actually catches something bad. Change-Id: Ia0021ca21538747855d725a2116d51b3b0480a1c Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31591Reviewed-by:
-
Ben Clayton authored
Bug: chromium:963658 Change-Id: I7db9bd4b5cec57c65298cdef512d18e31d4182f5 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31594 Presubmit-Ready: Ben Clayton <bclayton@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Nicolas Capens authored
Bug: b/129523279 Change-Id: I894b11a58fcde956312ad92f934aaa0c60a0f837 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/30808Tested-by:
-
Chris Forbes authored
Bug: b/117974925 Change-Id: I7a5cb6f8bbee13322e1d2524c9ecb8df20fe8e04 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31208Tested-by:
-
SwiftShader Regression Bot authored
Change-Id: Ib3b20d2268fc7db22817d2d073ecfec60eb1c30a Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31668 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
- 16 May, 2019 8 commits
-
-
Chris Forbes authored
- Renderer fence signaling only really worked for work in the same submission. If the fence was submitted standalone later, there would be no renderer work tagged with this fence, and so no ordering. - Get rid of old uncounted fence operations. This is a landmine. Bug: b/132897411 Test: dEQP-VK.api.copy_and_blit.* Test: dEQP-VK.binding_model.* Test: dEQP-VK.query_pool.* Test: dEQP-VK.api.object_management.* Test: dEQP-VK.memory.pipeline* Change-Id: I9050b96d111a6c348c5675724c8017c483c06182 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31188Reviewed-by:
-
Chris Forbes authored
Bug: b/118386749 Test: dEQP-VK.memory_model.* Change-Id: Ie0962df8e17264ec93265cd56b4a79e0d5f37c33 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31528Tested-by:
-
Chris Forbes authored
Test: dEQP-VK.memory_model.* Bug: b/127472316 Change-Id: I96447aba539e2ae9b4d42637fbe4a829f35093f5 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31513Tested-by:
-
Chris Forbes authored
Helps drive out the test bugs we're running into. Bug: b/132879304 Test: dEQP-VK.renderpass.suballocation.* Change-Id: I33e168a9bdc2385beff9f5d42e44f8632934febe Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31568Tested-by:
-
Chris Forbes authored
Previously we were using the layout from the pipeline, but that doesn't work if the pipeline is not bound yet. Conveniently, BindDescriptorSets is passed a pipeline pointer by the app for this reason. Bug: b/123244275 Test: dEQP-VK.binding_model.descriptorset_random.sets4.* Change-Id: I94616efa7f2acd92733df9f408714c5867a87f53 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31511Tested-by:
-
Chris Forbes authored
This doesn't behave as expected, and asserts inside LLVM. Test: dEQP-VK.binding_model.descriptorset_random.sets4.* Change-Id: I3a4d61a3dcac2c62d8e29982e0b7b2a82a22caa1 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31510Reviewed-by:
-
Ben Clayton authored
Bug: b/126873455 Tests: dEQP-VK.glsl.builtin.precision.ldexp.* Tests: dEQP-VK.glsl.builtin.function.common.ldexp.* Change-Id: If70125d749ba976abfa7295f067be51c5e8c2417 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31409 Presubmit-Ready: Ben Clayton <bclayton@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
SwiftShader Regression Bot authored
Change-Id: I06d65cdc8179c1cc15d3b38bf4d95143dd500825 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31588Reviewed-by:
-
- 15 May, 2019 10 commits
-
-
Hernan Liatis authored
This is just a small hack to prevent screen tearing. Later, we should properly implement VkSemaphore to prevent this. Bug: b/132458423 Change-Id: I825f100a73c4b5b9b909eba83552cbec51ff1792 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31152Reviewed-by:
-
Nicolas Capens authored
Change-Id: Iaf147233a0703360a2fb95a36aae5b2cf011dfc7 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31512Reviewed-by:
-
Chris Forbes authored
This was masked by the incorrect assert -- we did the wrong thing in MRT cases, but the assert intended to protect us would only fire if logic ops were also enabled. Fix this properly. We still only have one shared set of blend state, but each attachment has its masks set now. Some of the tests exercising this are dubious on a device which doesn't support independent blending -- the color write masks are technically part of the state which must be the same across all attachments. Bug: b/132433217 Test: dEQP-VK.renderpass.* Change-Id: Ia22f9cbf156a9f3c24d47142211bde67f06a0d46 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31350Tested-by:
-
Chris Forbes authored
SPIRV-Tools' pass to do this is incomplete, and implementing the remaining pieces as a SPIRV->SPIRV transform is more work than implementing them here. Related SPIRV-Tools issues: 2585 2586 Bug: b/127454276 Test: dEQP-VK.*quantize* Change-Id: I892de80dff366fb3bc417315b5db15d850577c47 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31348Tested-by:
-
Alexis Hetu authored
In order to easily add 32 bit support, this cl adds a header which essentially uses a union to make sure that even when we have 32 bit pointers, the Vulkan handles are 64 bit. This change should be noop. Bug b/129979580 b/127920555 Change-Id: I54254929186584ec4544a1da5a7def7cf56e392e Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31070 Presubmit-Ready: Alexis Hétu <sugoi@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Reviewed-by:
-
Chris Forbes authored
Vulkan always has separate alpha blend. No test change expected Bug: b/132280877 Change-Id: Ie76094fcc82070c20b78b846e4d2e30810339f98 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31349Tested-by:
-
Chris Forbes authored
Bug: b/119620767 Change-Id: If258773373426ed404cc396a672d4bc4078458d6 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31389Tested-by:
-
Chris Forbes authored
When we decompress ETC2 RGB into a 4-channel texture, *almost* everything just works as long as we ensure the alpha channel has the correct values -- except that to the sampler, this is now a 4-channel texture and so no channel replacement will be performed. This interacts badly with border texel replacement, which occurs before channel replacement. When a transparent border mode is used, we end up with non-1 values in the returned alpha channel, which is wrong. Instead, drive missing channels to 0/1 with the swizzle, which we control independently of the texture format. Bug: b/119620767 Test: dEQP-VK.pipeline.sampler.* Change-Id: I2816e18fc1c36bd4de51a3e3871c75c7dc1b8867 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31388Tested-by:
-
Ben Clayton authored
This is failing the Windows builds. There is no equivalent warning on Clang or GCC, and this seems harmless, so just silence it. Bug: b/132445520 Change-Id: I4c9498d87490fe892183e2d24c80979d785c37e8 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31428 Presubmit-Ready: Ben Clayton <bclayton@google.com> Reviewed-by:
Antonio Maiorano <amaiorano@google.com> Reviewed-by:
-
SwiftShader Regression Bot authored
Change-Id: I688c02f375ff9a1935cdd08f553afdc0c14e27a5 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31408Reviewed-by:
-
- 14 May, 2019 5 commits
-
-
Ben Clayton authored
Adds support for a2b10g10r10_unorm_pack32 render targets, and input attachments. Tests: dEQP-VK.renderpass.dedicated_allocation.formats.a2b10g10r10_unorm_pack32.* Tests: dEQP-VK.renderpass.suballocation.formats.a2b10g10r10_unorm_pack32.* Bug: b/131896622 Bug: b/131171141 Change-Id: I925723754858c6b3515d142d0c7abae275685d6d Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31271 Presubmit-Ready: Ben Clayton <bclayton@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Antonio Maiorano authored
* Respect WARNINGS_AS_ERRORS option for msvc builds. * Disable a bunch of warnings for msvc builds. These were gathered from all the hand-crafted vcxproj files and merged into one set. Eventually we need to disable some of these and fix the actual warnings. * Treat specific warnings as errors for msvc builds. These were gathered from the hand-crafted vcxproj files. * Fix LLVM build showing warning 4141 despite being disabled in CMakeLists. Something wonky happens when using set with the first string argument. Going with list(APPEND,...) works, and is symmetrical with the non-msvc script. * Define _SCL_SECURE_NO_WARNINGS for all modules, not just LLVM. * Fix double to float conversion warning in SprivShader.cpp * Fix mismatched type compare in IceTargetLoweringX664Traits.h Bug: b/132445520 Change-Id: I1143e9fdd386e26d4f0b0f9e34c4d6b3ac2f2311 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31149 Presubmit-Ready: Antonio Maiorano <amaiorano@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Ben Clayton authored
Loops attempted to special-case phis with allocas to deal with their non-ssa nature, but this didn't handle divergent loops correctly. Rearchitect phis to always use an alloca, and mask the updates with the edge masks. This handles all cases in a cleaner, more unified implementation. Tests: dEQP-VK.glsl.switch.* Tests: dEQP-VK.glsl.functions.control_flow.* Bug: b/128527271 Change-Id: I5b450cd3f5f10b1076d65750d326fedc286abfe5 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31269 Presubmit-Ready: Ben Clayton <bclayton@google.com> Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Tested-by:
-
Ben Clayton authored
Combine all the active lane masks for the merge block. Previously this was only using the lane mask of the last loop iteration. Loop phis were being overridden even for disabled lanes. Now fixed. Made EmitBranch() use addActiveLaneMaskEdge(). This is effectively a noop change, but keeps the code more consistent. dEQP-VK.glsl.switch.switch_in_do_while_loop_dynamic_vertex still seems broken, which I'm investigating. Tests: dEQP-VK.glsl.switch.* Bug: b/128527271 Change-Id: I49bafe5cdb49d6878dde98134f9cbb9a3ad4d384 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31268 Kokoro-Presubmit: kokoro <noreply+kokoro@google.com> Presubmit-Ready: Ben Clayton <bclayton@google.com> Tested-by:
-
SwiftShader Regression Bot authored
Change-Id: I6f90c060f388b932002d92b107473edf30e1f52a Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31270Reviewed-by:
-
- 13 May, 2019 2 commits
-
-
Chris Forbes authored
Tracks the first subpass to touch an attachment. Test: dEQP-VK.renderpass.suballocation.unused* Change-Id: Idc0998da022fd4c0b1785f6020c795916ba475f7 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/31308Tested-by:
-
Hernan Liatis authored
This is the minimum amount of code to get a triangle on an Android device using Swiftshader Bug: b/122837237 Change-Id: I8d2af6c104d70650c8750c9dd54d5be32ded0a39 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/30828Tested-by:
-
