kc3-lang/angle/src/libANGLE/renderer

src/libANGLE/renderer

Log

Author Commit Date CI Message
Shao dde78e8c 2017-05-22T14:13:27 ES31: Implement Vertex Attrib Binding on OpenGL This patch intends to implement Vertex Attrib Binding on OpenGL back-ends: 1. Add supports for updating vertex attributes by Vertex Attrib Binding APIs. 2. Refactor the process of updating vertex attribtues in class VertexArray to make it easier to implement this feature. BUG=angleproject:1593 TEST=dEQP-GLES31.functional.vertex_attribute_binding.* Change-Id: I800e61518c552b94b84c415895ad31668b0a84b2 Reviewed-on: https://chromium-review.googlesource.com/510251 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Geoff Lang c7ba85c9 2017-07-14T22:17:01 TexutreD3D: Only update storage level if it exists. This was causing an assertion failure in updateStorageLevel when the storage didn't contain the requested level. Gate the call behind an isValidLevel check. TEST=conformance/misc/type-conversion-test.html BUG=angleproject:1815 Change-Id: Iadf46d232d5a117ae1ae74c6e71677b1e06a5321 Reviewed-on: https://chromium-review.googlesource.com/572705 Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Geoff Lang <geofflang@chromium.org>
Martin Radev 318f9aa5 2017-05-17T17:47:28 Initialize ANGLE_multiview caps and workaround state The patch checks whether ANGLE_multiview can be supported in the OpenGL renderer, updates the caps and adds a workaround field to enable multiview support through the NV_viewport_array2 extension. BUG=angleproject:2062 TEST=angle_end2end_tests Change-Id: I99dae10564db7bcca41d7624f8de272c1d996e09 Reviewed-on: https://chromium-review.googlesource.com/567934 Reviewed-by: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Geoff Lang c5af8ba6 2017-07-11T12:18:31 D3D11: Make sure to resolve the storage for CopyTexImage3D. TextureD3D_3D::copySubImage worked around missing functionality in D3D11 for copying a framebuffer directly to a texture storage but didn't handle the case of a texture storage already existing. This caused the image to have out-of-date data before the new data was copied into it. Simply copy the data from the storage back into the image before performing the copy from the framebuffer and then copy back to the storage afterwards. TEST=conformance2/textures/misc/copy-texture-image-webgl-specific.html BUG=angleproject:1815 Change-Id: I308d6a1d3ecbc738f7d0e232bece433e6b353638 Reviewed-on: https://chromium-review.googlesource.com/567199 Commit-Queue: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Geoff Lang d90b214f 2017-07-12T15:33:35 D3D11: If an image is dirty, copy it to the storage before copying from a FB. If a storage already existed, it may have old data in it. Check if the image is dirty and copy the initialized memory into the storage before copying data from the framebuffer. TEST=conformance/textures/misc/copy-tex-image-and-sub-image-2d BUG=angleproject:1815 Change-Id: Ic69c5519b2e09e4b62025b1bf413d2a71a4a2afb Reviewed-on: https://chromium-review.googlesource.com/568410 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Geoff Lang <geofflang@chromium.org>
JiangYizhou 34bc315d 2017-03-29T14:56:01 ES31: Implement multisampled Textures for D3D part. Implement TexStorage2DMultisample api for d3d part. BUG=angleproject:1590 TEST=angle_end2end_tests --gtest_filter=TextureMultisampleTest* TEST=angle_deqp_gles31_tests --deqp-case=dEQP-GLES31.functional.texture.multisample.negative.fbo_* Change-Id: Icbfba45b9c2965af02b54dd4060b7b49970cb74b Reviewed-on: https://chromium-review.googlesource.com/457161 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit Bot 84db5733 2017-07-13T08:00:02 Merge "Fix non statically used fragment input structs on HLSL"
Commit Bot 1636e1b9 2017-07-12T14:29:06 Merge "D3D11: Clip copy rect to the source framebuffer for copyTexImage3D."
Olli Etuaho 06a06f5e 2017-07-12T12:22:15 Fix non statically used fragment input structs on HLSL Add static use information to struct fields that mirrors the static use information on the struct itself. This way dynamically generated HLSL doesn't need special handling for initializing fragment inputs if they are structs. This fixes a problem with the previous code where dynamically generated HLSL ended up trying to initialize structs that are not declared in the HLSL output because they were not being referenced. BUG=angleproject:2104 TEST=angle_end2end_tests Change-Id: I21283ce4fe26515d62d95e61f8155dc9a9b44cf1
Geoff Lang bc5d7add 2017-07-10T16:17:26 D3D11: Clip copy rect to the source framebuffer for copyTexImage3D. TEST=conformance2/textures/misc/copy-texture-image-webgl-specific.html BUG=angleproject:1815 Change-Id: I146fcf97a9c90f07d6270672c5e44e05602eecf8
Brandon Jones 340b7b8b 2017-06-26T13:02:31 Add support for new internalformats in copyTextureCHROMIUM This adds support in blit11::copyTexture for LUMA, LUMA_ALPHA, and ALPHA formats as destinations. Added is handling for each case to match up the corresponding shader. This required new premultiply and unmultiply D3D11 shaders for some cases. Changed copyTextureCHROMIUM validation to allow new formats. Tests have been created to demonstrate using copyTextureCHROMIUM with the new formats with default parameters, as well as with the unpackPremultiply and unpackUnmultiply parameters. BUG=:angleproject:2101 Change-Id: Id8cd303a46fe70710bc18172fc938552a6e4cfaf
Geoff Lang 88318b44 2017-07-05T14:39:01 Fall back to CPU copies for srgb textures in copy_texture_CHROMIUM. The copied data is not supposed to have sRGB conversions applied to it when written to the destination texture but an sRGB SRV is used by Blit11. Instead of creating multiple sRGB and non-sRGB SRVs for textures, simply fall back to the CPU copy path for this format for now. Clip color channels that should not exist in the destination texture formats in Image11::CopyImage. This works around issues with texture formats with emulated channels. TEST=conformance2/textures/image_bitmap_from_canvas/tex-2d-srgb8-rgb-unsigned_byte TEST=conformance2/textures/image_bitmap_from_canvas/tex-2d-srgb8_alpha8-rgba-unsigned_byte BUG=angleproject:1932 Change-Id: Ieeda3569f80d016fda781e7eb498acd3b97568d0 Reviewed-on: https://chromium-review.googlesource.com/559857 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Geoff Lang <geofflang@chromium.org>
Bryan Bernhart (Intel Americas Inc) caf20889 2017-07-05T15:26:44 Fix incorrect quality level with depth-stencil buffer and MSAA configs. Using a depth buffer format requires that the DSV and RTV have equal quality levels; otherwise, the bound render target will discard writes. BUG=angleproject:1917 Change-Id: Ife25b0a8958fa2b31b43a0d877d27e440916a9bf Reviewed-on: https://chromium-review.googlesource.com/560716 Reviewed-by: Bryan Bernhart <bryan.bernhart@intel.com> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Geoff Lang <geofflang@chromium.org>
Olli Etuaho a3e1f5f7 2017-07-06T13:21:07 Fix determining max combined uniform blocks The previous code was using a completely unrelated variable to compute max combined uniform blocks limit. BUG=angleproject:2099 Change-Id: I9e56b50a92790f525dda50adca52b6ac5edfc95a Reviewed-on: https://chromium-review.googlesource.com/562276 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Frank Henigman 1cbe9203 2017-06-15T18:13:31 Clip TextureGL::copyImage to framebuffer. Ensure the underlying GL does not modify areas of the texture that correspond to areas outside the framebuffer, as required for WebGL. Also zero out the texture in WebGL mode because CopyTexImage must return zeroes for areas outside the framebuffer. Enable corresponding test. BUG=angleproject:1815 Change-Id: I51b1221dbf0dda0952e2ae89ee6ac925b5d1d4a4 Reviewed-on: https://chromium-review.googlesource.com/551535 Commit-Queue: Frank Henigman <fjhenigman@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Frank Henigman 95ba174e 2017-06-22T20:38:04 Clip TextureD3D_2D::copyImage to framebuffer. WebGL CopyTexImage needs to zero the part of the texture corresponding to area outside the framebuffer, so we zero the whole texture then clip the read area. The clipping also avoids problems with code lower down that isn't prepared for read areas not entirely within the framebuffer. Enable corresponding test. BUG=angleproject:1815 Change-Id: Ia7e0243ca72fa7c8f5bacda4d2022061d6a6d4f0 Reviewed-on: https://chromium-review.googlesource.com/551056 Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Frank Henigman <fjhenigman@chromium.org>
Frank Henigman fa36c330 2017-06-09T18:44:45 Clip FramebufferGL::readPixels to framebuffer. In GL, ReadPixels() is allowed to modify memory that corresponds to pixels outside the framebuffer. In WebGL it must not do that, so clip the read area to the framebuffer. Enable corresponding test. BUG=angleproject:1815 Change-Id: I8113ae417dee7834e63498aec8291ce711bd7513 Reviewed-on: https://chromium-review.googlesource.com/536434 Commit-Queue: Frank Henigman <fjhenigman@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 50c562de 2017-06-06T14:43:30 Re-land: Don't expose non-conformant multisampling modes on GL Re-landing with a fallback for failed internal format queries to work around issue seen on Shield TV. Also fixed wrong handling of integer RG formats in isRequiredRenderbufferFormat. Some NVIDIA GL drivers expose non-conformant multisampling modes. The conformance of multisampling modes can be queried using the extension NV_internalformat_sample_query. Use it to filter out the non-conformant modes from the modes that are exposed by ANGLE. The MAX_SAMPLES value and other similar values stored in caps also need to be lowered to match the maximum number of samples exposed for required formats. There seems to be an NVIDIA driver bug related to querying STENCIL_INDEX8 multisample format. Work around this by querying DEPTH24_STENCIL8 instead. There's also some confusion around whether RGB9_E5 should be renderable. Once the floating point texture extensions got rolled into the core GL spec, it was eventually made clear that RGB9_E5 is intended not to be renderable. The extension specs that predate float textures in the core spec do suggest that it would be renderable, but in practice drivers that advertise the extension strings don't reliably implement RGB9_E5 as renderable. Solve this by disabling it as a renderable format and adding an explanatory comment. BUG=chromium:682815 TEST=angle_end2end_tests, dEQP-GLES31.functional.state_query.internal_format.renderbuffer.* Change-Id: I727f03045a1534d6764b571e6d839243705d25b3 Reviewed-on: https://chromium-review.googlesource.com/551957 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 27a60631 2017-06-30T15:12:01 Re-apply UBO binding workaround on program save. The workaround which was previously defined to only apply on load also seems to affect save on some AMD drivers. BUG=angleproject:1637 BUG=angleproject:1897 Change-Id: Ia01a1420a484f3c2682ce97eaab18baccfb66a50 Reviewed-on: https://chromium-review.googlesource.com/558008 Reviewed-by: Yuly Novikov <ynovikov@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Frank Henigman 739bd8ba 2017-06-19T21:02:27 Clip FramebufferD3D::readPixels to framebuffer. In GL, ReadPixels() is allowed to modify memory that corresponds to pixels outside the framebuffer. In WebGL it must not do that, so clip the read area to the framebuffer. Enable corresponding test. BUG=angleproject:1815 Change-Id: Ie99401a2102c352ffb1193a57aa66a5b96c184aa Reviewed-on: https://chromium-review.googlesource.com/540556 Commit-Queue: Frank Henigman <fjhenigman@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org>
jchen10 56375021 2017-06-21T11:38:04 D3D11: Fix Buffer11 for GPU-to-CPU storage copy. When copying from a non-mappable NativeStorage to a system memory based storage, a staging buffer storage should be used. Currently this is only done for PackStorage. This covers the missed SystemMemoryStorage and EmulatedIndexedStorage. Also it adds a triggering test case to expose the bug. BUG=angleproject:2076 Change-Id: I278a0eef85751e966c1c48ddd71010092a14a3f7 Reviewed-on: https://chromium-review.googlesource.com/542595 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Frank Henigman ab04e6a7 2017-06-22T20:38:04 Clip TextureD3D_2D::copySubImage to framebuffer. WebGL CopyTexSubImage does not allow touching parts of the texture that correspond to area outside the framebuffer, so we clip the read area to the framebuffer. The clipping also avoids problems with code lower down that isn't prepared for read areas not entirely within the framebuffer. Enable corresponding test. BUG=angleproject:1815 Change-Id: I411223669dae2a456dfc3e22acda907b73177988 Reviewed-on: https://chromium-review.googlesource.com/527411 Commit-Queue: Frank Henigman <fjhenigman@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Frank Henigman 79fd1e9f 2017-06-15T18:13:31 Clip TextureGL::copySubImage to framebuffer. Ensure the underlying GL does not modify areas of the texture that correspond to areas outside the framebuffer, as required for WebGL. Enable corresponding test. BUG=angleproject:1815 Change-Id: I6092d39e43868902de7ae3aee430deea3b3ff8a1 Reviewed-on: https://chromium-review.googlesource.com/538295 Commit-Queue: Frank Henigman <fjhenigman@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Yuly Novikov c8a8b843 2017-06-28T01:16:41 Revert "Don't expose non-conformant multisampling modes on GL" This reverts commit 3cd0dd370f08e7126fe19b761206280c64f80cd1. Reason for revert: crashes on NVIDIA Shield TV https://build.chromium.org/p/chromium.gpu.fyi/builders/Android%20Release%20%28NVIDIA%20Shield%20TV%29/builds/1816 Original change's description: > Don't expose non-conformant multisampling modes on GL > > Some NVIDIA GL drivers expose non-conformant multisampling modes. The > conformance of multisampling modes can be queried using the extension > NV_internalformat_sample_query. Use it to filter out the > non-conformant modes from the modes that are exposed by ANGLE. > > The MAX_SAMPLES value and other similar values stored in caps also > need to be lowered to match the maximum number of samples exposed > for required formats. > > There seems to be an NVIDIA driver bug related to querying > STENCIL_INDEX8 multisample format. Work around this by querying > DEPTH24_STENCIL8 instead. > > There's also some confusion around whether RGB9_E5 should be > renderable. Once the floating point texture extensions got rolled > into the core GL spec, it was eventually made clear that RGB9_E5 > is intended not to be renderable. The extension specs that predate > float textures in the core spec do suggest that it would be > renderable, but in practice drivers that advertise the extension > strings don't reliably implement RGB9_E5 as renderable. Solve this > by disabling it as a renderable format and adding an explanatory > comment. > > BUG=chromium:682815 > TEST=angle_end2end_tests, > dEQP-GLES31.functional.state_query.internal_format.renderbuffer.* > > Change-Id: I2218e3a23ea7b48a0615fea77a91897dc7d5fe9e > Reviewed-on: https://chromium-review.googlesource.com/525515 > Reviewed-by: Geoff Lang <geofflang@chromium.org> > Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> TBR=geofflang@chromium.org,jmadill@chromium.org,cwallez@chromium.org,oetuaho@nvidia.com # Not skipping CQ checks because original CL landed > 1 day ago. Bug: chromium:682815 Change-Id: I1ebdf52c3fab1526f5a561ac4c8555e305ef2243 Reviewed-on: https://chromium-review.googlesource.com/551164 Reviewed-by: Yuly Novikov <ynovikov@chromium.org> Commit-Queue: Yuly Novikov <ynovikov@chromium.org>
Jamie Madill ffe00c03 2017-06-27T16:26:55 Add program cache transform feedback workaround. On Qualcomm devices, they don't seem to correctly save transform feedback info. Work around this by disabling caching on these devices. This mirrors a Chromium workaround. BUG=angleproject:2088 Change-Id: I6496d2fb6a03788379a6968bcd5eb3a9cb9d15d4 Reviewed-on: https://chromium-review.googlesource.com/549981 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Jamie Madill 761b02c8 2017-06-23T16:27:06 Add an applyNativeWorkarounds context impl hook. This method can allow the implementation to override the Context's workarounds. Use this design pattern now that we have access to the gl::Context everywhere - we don't need to cache a local copy in the Renderer objects. This will be used to apply a Shader Program Cache workaround on the GL level, that will only be used for the GLES back-end on Qualcomm. BUG=angleproject:2088 Change-Id: I6da25c5c29c3ba01b8820c5234d1b92dd2d2121a Reviewed-on: https://chromium-review.googlesource.com/549980 Reviewed-by: Yuly Novikov <ynovikov@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill aed7c570 2017-06-23T14:18:36 GL: Don't call seamless cube cap on GLES. This cap is only valid on desktop GL. There might be extensions where the feature is available as well, so we should follow up and check for the extensions as well. This fixes a Debug runtime warning in almost every end2end_test on GLES. BUG=angleproject:2085 Change-Id: I5edc1c667b58230df903da82de2a8aceb0369c0c Reviewed-on: https://chromium-review.googlesource.com/546597 Reviewed-by: Yuly Novikov <ynovikov@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill c6a34b03 2017-06-23T10:51:00 Add guards around dxgi_support_table.h. BUG=angleproject:2082 Change-Id: I6f20502bac3c664d5ef8c6d65b244c37a5727f26 Reviewed-on: https://chromium-review.googlesource.com/545458 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org> Commit-Queue: Geoff Lang <geofflang@chromium.org>
Corentin Wallez 84a6c5b2 2017-06-21T16:12:58 Perform all glXMakeCurrent in DisplayGLX This allows caching the current surface more efficiently than at the egl layer because we can take advantage of having only one backing context. BUG=angleproject:1651 Change-Id: I62867b16ac5e06901a988dc41a3d4812accdb74c Reviewed-on: https://chromium-review.googlesource.com/543835 Commit-Queue: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Geoff Lang 665e4d97 2017-06-21T15:04:48 Perform all wglMakeCurrent calls from DisplayWGL. Display WGL is the best place to track the currently bound surfaces and contexts and cache them appropriately. This results in a ~40% speed increase in the WebGL Aqarium demo because wglMakeCurrent is now only called once at initialization. BUG=angleproject:1651 Change-Id: I61132bb4a0db8de592600173c20787c67d18e067 Reviewed-on: https://chromium-review.googlesource.com/543583 Commit-Queue: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Xinghua Cao 65ec0b2e 2017-03-28T16:10:52 ES31: Add support for bindImageTexture on GL backend This patch refers to https://chromium-review.googlesource.com/c/380636/ BUG=angleproject:1987 Change-Id: If621eed6ecaa7298214843a2a133801ca1487b03 Reviewed-on: https://chromium-review.googlesource.com/462088 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
jchen10 7a20b973 2017-06-13T14:25:26 Refactor uniform block binding Remove mUniformBlockBindings and move its bindings to mUniformBlocks. BUG=angleproject:1442 Change-Id: I62b4471990a44e626d2357c41cb914abc27cb18f Reviewed-on: https://chromium-review.googlesource.com/532834 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Corentin Wallez db9e5d31 2017-06-12T12:05:45 D3D11: Only apply attachments that are written by the program This works around a bug in the AMD driver that writes 0's to the first attachment if it isn't written by the pixel shader. BUG=angleproject:2048 Change-Id: I384fd60c0e0a37fbc0fd7b69fe1ec74fe4ffac8f Reviewed-on: https://chromium-review.googlesource.com/531630 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Geoff Lang 83dde7b9 2017-06-21T15:02:54 Use the std::unordered_map::at method instead of operator[]. It is known that the entry exists in the map, the at operator is slightly faster in this case. BUG=angleproject:2079 Change-Id: I3a64212293fa0320f3df5327e92d11eb7ef28c4f Reviewed-on: https://chromium-review.googlesource.com/543683 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Geoff Lang <geofflang@chromium.org>
Jamie Madill da066657 2017-06-21T11:33:48 Fix invalid storage recovery on FL 9_3. This problem was uncovered when I inadvertently changed TextureD3D to delete its storage before its images. Small mips of compressed textures must use a nullptr argument to CopySubResource, otherwise the runtime complains about un-aligned sizes. Also change the class to delete the Images before the Storage again so we don't wastefully recover the images before deleting them. Also change the Image pointers to use std::array and std::unique_ptr. BUG=angleproject:1156 BUG=angleproject:2077 Change-Id: Idb2e53835b7a9b973285ff0781f70b25f05c77aa Reviewed-on: https://chromium-review.googlesource.com/543438 Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 4928b7ca 2017-06-20T12:57:39 Proliferate gl::Context everywhere. This gives the D3D back-end access to the GL state almost anywhere. This uses the onDestroy hook for Textures to push errors up from destructors, although they still don't quite make it to the Context. There are places, such as in EGL object (Context/Surface) destruction, where we end up calling through to GL implementation internals without having access to a gl::Context. We handle this via a proxy Context to a Display, basically a null context, that has access to impl-side state like the Renderer pointer if necessary. It does not have access to the normal GL state. Also Pass gl::Context to RefCountObject::release(). Since we're using destroy() methods now, we should not ever call the destructor directly. BUG=angleproject:1156 Change-Id: Ie4c32ad6bf6caaff0289901f30b5c6bafa2ce259 Reviewed-on: https://chromium-review.googlesource.com/529707 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Bruce Dawson 6260b7aa 2017-06-19T11:07:19 Fix kResourceTypeNames initialization Due to a missing comma the kResourceTypeNames array contained a single long (concatenated) string instead of NumResourceTypes independent strings. This incorrect initialization caused a crash in out-of-memory situations, but this was only noticed on VC++ 2017 for some reason. This fix adds the missing comma and uses a static_assert to ensure that the array is initialized correctly. BUG=chromium:727671,728226,731089 Change-Id: I9f0f3d3725b9f773505506513afb6c349db3a7fb Reviewed-on: https://chromium-review.googlesource.com/539536 Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 3cd0dd37 2017-06-06T14:43:30 Don't expose non-conformant multisampling modes on GL Some NVIDIA GL drivers expose non-conformant multisampling modes. The conformance of multisampling modes can be queried using the extension NV_internalformat_sample_query. Use it to filter out the non-conformant modes from the modes that are exposed by ANGLE. The MAX_SAMPLES value and other similar values stored in caps also need to be lowered to match the maximum number of samples exposed for required formats. There seems to be an NVIDIA driver bug related to querying STENCIL_INDEX8 multisample format. Work around this by querying DEPTH24_STENCIL8 instead. There's also some confusion around whether RGB9_E5 should be renderable. Once the floating point texture extensions got rolled into the core GL spec, it was eventually made clear that RGB9_E5 is intended not to be renderable. The extension specs that predate float textures in the core spec do suggest that it would be renderable, but in practice drivers that advertise the extension strings don't reliably implement RGB9_E5 as renderable. Solve this by disabling it as a renderable format and adding an explanatory comment. BUG=chromium:682815 TEST=angle_end2end_tests, dEQP-GLES31.functional.state_query.internal_format.renderbuffer.* Change-Id: I2218e3a23ea7b48a0615fea77a91897dc7d5fe9e Reviewed-on: https://chromium-review.googlesource.com/525515 Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Jamie Madill 30ca54f4 2017-06-19T11:45:06 D3D11: Fix mingw64 build. Two small build errors were breaking ming64. BUG=angleproject:2071 Change-Id: Ia5c8e629e77c09f151b888364e92475b4c3f1709 Reviewed-on: https://chromium-review.googlesource.com/539796 Reviewed-by: Jamie Madill <jmadill@chromium.org>
Geoff Lang 125e220e 2017-06-19T13:23:18 Track LevelInfoGL per cube map face. When transitioning between formats in cube maps, tracking of luma workaround information would become incorrect because it was only tracked per-level. There are no failing tests due to this because once the cube map is complete again, the tracking is correct. TEST=conformance/more/conformance/quickCheckAPI-S_V.html BUG=angleproject:2074 Change-Id: If4e1e94061208f0cf3668b2bd6be59a9a21e42b6 Reviewed-on: https://chromium-review.googlesource.com/539956 Commit-Queue: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Jamie Madill fe54834f 2017-06-19T11:13:24 Proliferate gl::Context. This enables a few small things: it will enable making the platform a property of the Display rather than a global. The same goes for the global logging annotator. Also it ensures all back-end implementations have access to the GL / EGL state when available. Also introduces a smart pointer helper class to angleutils for objects that prefer to be destroyed with a context (gl::Context/egl::Display) parameter. We were using std::unique_ptr in a few places that would not work well with these objects. BUG=angleproject:1156 Change-Id: I59e288a3d6f766ff8a0f4b48ff3a1fbf7489daba Reviewed-on: https://chromium-review.googlesource.com/529706 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Geoff Lang 638c7277 2017-06-13T15:50:26 Implement EGL_KHR_surfaceless_context for D3D, GL and NULL backends. Skip all config attributes that have the value of EGL_DONT_CARE. From the EGL 1.5 spec: "If EGL_DONT_CARE is specified as an attribute value, then the attribute will not be checked.". BUG=angleproject:1651 Change-Id: I30c95a1970543fb6f1d4b02d2babf3df61cad543 Reviewed-on: https://chromium-review.googlesource.com/533937 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Geoff Lang <geofflang@chromium.org>
Geoff Lang 165dcf15 2017-06-07T15:05:14 Allow CHROMIUM_copy_texture to copy to non-zero mips. Fix some errors in the GL backend related to source and destination mipmap targets. BUG=angleproject:1356 Change-Id: I030529c8626f3bc30dbb4f7f859a02ba56e315e3 Reviewed-on: https://chromium-review.googlesource.com/527653 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Geoff Lang <geofflang@chromium.org>
Jamie Madill 97ee6548 2017-06-07T17:06:05 Shunt more code to StateManager11. Previously the Renderer11 would call directly into the state manager sync methods. Instead make a single updateState method, and make several state sync methods private to the manager. Also rename them to clarify they're for syncing state, not for direct use. BUG=angleproject:1156 Change-Id: I94880a744e7ade3895fa2a312a2436ba4ef38dba Reviewed-on: https://chromium-review.googlesource.com/529705 Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill fb05bcba 2017-06-07T15:43:18 Move the vk Serial class to renderer_utils. This can be useful for other back-ends, for various types of state management. Also redesign the class to use an opaque factory instead of an increment operator. The class maintains the property of being ordered. Also assume we don't overflow with 64-bit serials. We could maybe redesign this to use 32-bit serials for memory constrained situations, and handle overflow more gracefully. I plan to use the serials to track state revisions for the vertex array class, to avoid doing redundant work. BUG=angleproject:1156 Change-Id: I02c78b228bc6e2fb3ee786fe67a4e607baaca18e Reviewed-on: https://chromium-review.googlesource.com/529704 Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill c174db3a 2017-06-13T15:44:00 D3D11: Apply a non-null blendstate in masked depth clear. In https://chromium-review.googlesource.com/c/453888/, we changed how blendstates were applied in masked clears. This change would apply a null blendstate when all color channels were disabled, but this seems to have an issue on Intel Cherry View drivers. Work around this issue by restoring the prior functionality of making a simple masked blend state. Also clean up some of the code style in the Clear11 class. BUG=chromium:730126 Change-Id: I9a4044201b2f07e9483525513a59e19bb2a8bcd3 Reviewed-on: https://chromium-review.googlesource.com/533684 Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill d222d454 2017-06-06T17:48:35 Use HashingMRUCache in InputLayoutCache. This simplifies the state management code inside this manager. Also it should improve the speed of lookups due to using hashing instead of a tree lookup. BUG=angleproject:1156 BUG=angleproject:2044 Change-Id: I19ea8dbac6f2dfd7d30dd403d77b66ba0aa85d73 Reviewed-on: https://chromium-review.googlesource.com/527693 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Jamie Madill 34ca4f5b 2017-06-13T11:49:39 Remove Shader::getSemanticIndex. This method doesn't really belong to GL, and was only used by the D3D back-end to compute some attribute indexes. Simplify the code by moving it into ProgramD3D. Also add the ability for the ShaderImpl to assert that any pending compiles have resolved. BUG=angleproject:1156 Change-Id: I0af3d3082ff8c908e6a87b9734989efbefd28808 Reviewed-on: https://chromium-review.googlesource.com/526336 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Martin Radev dd5f27ee 2017-06-07T10:17:09 Make VertexBinding's member variables private The patch decorates all members in VertexBinding as private and limits access to them only through getters and setters. This makes it easier to debug and keep track of any assignments to the class members. BUG=angleproject:2062 TEST=angle_end2end_tests Change-Id: Iddd49063d060f136bc9cf11c313a5af0931d433c Reviewed-on: https://chromium-review.googlesource.com/530786 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Jamie Madill 9cf9e871 2017-06-05T12:59:25 Move LinkResult to the gl:: namespace. This is a derived type from gl::Error so makes sense to be there. Also makes it more accessible than in ProgramImpl.h BUG=angleproject:1897 Change-Id: Id41b13e5a072745d8c361057f5bef8f152e0452b Reviewed-on: https://chromium-review.googlesource.com/522872 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 83418fb1 2017-06-05T12:59:24 Don't allow for error in ProgramImpl::save. Refactoring cleanup patch only. BUG=angleproject:1897 Change-Id: I6d12de5dab16ead9684886a1cf15b570e3c98156 Reviewed-on: https://chromium-review.googlesource.com/522871 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Corentin Wallez c084de14 2017-06-05T14:28:52 ProgramD3D: only broadcast when the GL_EXT_draw_buffers is enabled The behavior is undefined in GL ES when multiple attachments are used with gl_FragColor but WebGL clarifies that: - If the extension is enabled, broadcast happens - If the extension isn't enabled, only the first attachment is written BUG=angleproject:2048 Change-Id: I6d85ba91df77d42fb8863a03f1faa006cd7817bf Reviewed-on: https://chromium-review.googlesource.com/523809 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Kai Ninomiya 0492d447 2017-06-07T13:45:15 Move murmurhash to src/common/third_party Bug: chromium:697758 Change-Id: I8a3a990b14cde0fdd45319d593040bfc571abf3e Reviewed-on: https://chromium-review.googlesource.com/527602 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Frank Henigman 60f6eb20 2017-05-08T15:34:46 Add WebGL test that draws to missing attachment. In WebGL one can Clear, DrawArrays, and DrawElements to a framebuffer with a missing attachment with no error. BUG=angleproject:1822 Change-Id: I4dece2fa8fad31c812e24ae18bdc380c2857a1f8 Reviewed-on: https://chromium-review.googlesource.com/502967 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Frank Henigman <fjhenigman@chromium.org>
Jamie Madill 2cd45629 2017-06-08T20:14:13 D3D11: Consolidate Vertex Buffer application. With this change, IASetVertexBuffers is only called from a single code site in StateManager11. All other classes call through to here. This will make adopting the dirty bits for InputLayouts and VBs much simpler. BUG=angleproject:2052 Change-Id: I6e7b6146deb7f80f5f0e75dd0aff18eb4cc2bfc2 Reviewed-on: https://chromium-review.googlesource.com/524232 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Geoff Lang 863b6236 2017-06-08T11:18:49 Refactor redefineImage and track dirty images properly. Several issues showed up in testing with WebGL: * Images should only be forcefully re-defined when there is no data to upload. * After an image is marked dirty, a later call to subImage would cause assertion failures because the texture storage would try to verify that the image was not dirty, don't try to copy directly to storage in this case. BUG=angleproject:1635 Change-Id: I9e5d83850d743b7d4d2db938312ee5c35a3a79ee Reviewed-on: https://chromium-review.googlesource.com/527348 Commit-Queue: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill 982f6e01 2017-06-07T14:33:04 Improvements to the gl::Range class. Make this a proper class, fix the extends method (previously did not work as expected), add a contains method, and add tests. Also add an iterator helper class so we can iterate over the range with range-for loops. This also fixes the shader resource unsetting code, which was not actually unsetting all the possible applied textures. BUG=angleproject:2052 Change-Id: I2a6fa97f96ccb612ad01a5e3f24dc869c54c967b Reviewed-on: https://chromium-review.googlesource.com/527318 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 5ea762a6 2017-06-07T14:59:51 Consolidate base:: imports. This merges the mrucache and numerics into one folder, to prevent having to make two identical versions of logging. BUG=angleproject:2044 Change-Id: Iba6dab05b21eb9ba0de44f27a90579c590a9a7fd Reviewed-on: https://chromium-review.googlesource.com/522870 Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Yuly Novikov c4d18aac 2017-03-09T18:45:02 Use ErrorStream everywhere Eliminates one more usage of FormatString and its static initializer. Add more ErrorStream types and replace gl::Error and egl::Error with them. BUG=angleproject:1644 Change-Id: Ib498d0ae4b81a332ec71aed7cf709993b154e6bb Reviewed-on: https://chromium-review.googlesource.com/505429 Commit-Queue: Yuly Novikov <ynovikov@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Geoff Lang 79460702 2017-06-06T15:30:45 Don't discard the storage on forceRelease, simply mark the image as dirty. It was overkill to always destroy the storage, especially for cube maps. Instead, just make sure the image thinks it's dirty when robust resource init is enabled so that it is always copied to the storage after redefinition. BUG=angleproject:1635 Change-Id: Ieba065d1873ca66a216bd05f4f8afa9c75349ff9 Reviewed-on: https://chromium-review.googlesource.com/526295 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Geoff Lang <geofflang@chromium.org>
Jamie Madill 966cb898 2017-06-05T16:08:27 D3D11: Don't dirty all uniforms on program change. (2/2) Second patch for addressing missed comments. BUG=angleproject:2053 Change-Id: I1268d571913f8ce7efa7b94a5d4e47cf9a80c3c0 Reviewed-on: https://chromium-review.googlesource.com/526275 Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill dcc9b51d 2017-06-05T15:28:45 D3D11: Consolidate input layout application. This merges all calls to IASetInputLayout to a single place in StateManager11. This means we no longer have to invalidate the state for D3D11, and can always lazily apply the input layout state. Introduces a new ResourceSerial class to replace the uintptr_t and DirtyPointer design. BUG=angleproject:2052 Change-Id: I76b874218b754395f25a129967c769b1f8f82115 Reviewed-on: https://chromium-review.googlesource.com/523025 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 07b36365 2017-06-05T16:08:27 D3D11: Don't dirty all uniforms on program change. Only dirty the uniforms if the uniforms are dirty. This gives about a 20% reduction in ANGLE overhead on a targeted micro-benchmark. It probably won't have that large of an impact on most real-world applications, but it was strictly unnecessary work. Something that came up when investigating the internal program binary cache in ANGLE D3D11. BUG=angleproject:2053 Change-Id: I2d55c010c29ce9627a9001dd2abe6e549f0b7e8c Reviewed-on: https://chromium-review.googlesource.com/524297 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 10bed9fc 2017-06-05T12:59:22 Minor optimizations to DynamicHLSL. This makes us use std::ostringstream in more places, instead of string concatenation. BUG=chromium:697758 Change-Id: Ifdcaa2e7e119664fc9cfdc566ea13b519a294714 Reviewed-on: https://chromium-review.googlesource.com/521729 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Geoff Lang f2a06035 2017-06-05T14:07:36 Force-release images on redefine when robust resource init is enabled. If the user repreatedly calls glTexImage with null data and the same size, it is expected that the texture would re-fill itself with zero'd data. TextureD3D and ImageD3D would no-op these calls when the texture is already the right size. TEST=conformance2/rendering/blitframebuffer-filter-outofbounds BUG=angleproject:1635 Change-Id: I810b15922759a4acfc0ef7da198f068b81e9efc4 Reviewed-on: https://chromium-review.googlesource.com/524436 Commit-Queue: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Geoff Lang 93a540fc 2017-06-05T14:26:47 D3D11: Reset the TF binding offsets when glBeginTransformFeedback is called. BUG=angleproject:2051 Change-Id: I09e8548ef76b7d824743d06b0bba21633bc40a24 Reviewed-on: https://chromium-review.googlesource.com/523671 Commit-Queue: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Jamie Madill bd044ed8 2017-06-05T12:59:21 Defer shader compiles when possible. When using the program binary memory cache inside ANGLE, this will give a potential fast path. If the user doesn't query the shader compile status or info log before calling LinkProgram, then we can check the program cache before translating the program, and if it finds a hit, we don't even need to call the translator. To preserve the shader settings at compile time, a reference to the current shader translator is kept in a binding pointer on the call to compile. This mirrors a similar implementation in Chromium's command buffer. Also the compile options and source are cached at compile to preserve the correct shader state. BUG=angleproject:1897 Change-Id: I3c046d7ac8c3b5c8cc169c4802ffe47f95537212 Reviewed-on: https://chromium-review.googlesource.com/517379 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Corentin Wallez 80cdc376 2017-04-18T13:27:30 DisplayWGL: try all GL context version from the highest The only way to get a core profile context of the highest version using CreateContextAttrib is to try creationg contexts in decreasing version numbers. It might look that asking for a core context of version (0, 0) works on some driver but it create a _compatibility_ context of the highest version instead which might not be supported on some drivers. BUG=angleproject:1994 Change-Id: Ic8a33243909c54ea5940ee81c9c866b141bea420 Reviewed-on: https://chromium-review.googlesource.com/480739 Commit-Queue: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill 63fa2e94 2017-06-01T17:11:41 D3D11: Consolidate Query allocation. This compiles the resource refactor for D3D11. The only remaining non-smart handles are for things like SwapChains and ID3D11Blob. This might be more easily managed by using angle::ComPtr or similar smart pointers instead of the Resource Manager class. BUG=angleproject:2034 Change-Id: I53c753a1c19b05c4e72859dd5046e3ed10866847 Reviewed-on: https://chromium-review.googlesource.com/506777 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 0fd806f1 2017-06-01T17:11:40 D3D11: Consolidate Shader allocation. Similar to the InputLayout init, this adds a small helper type to act as a wrapper around shader init data (binary and size). This also adds error trapping to the blit shader compilation. It also removes the LazyResource2 class, and the CompileXS helper methods. BUG=angleproject:2034 Change-Id: I3fd718393c8a0250e4263890f00d0e9147ec9567 Reviewed-on: https://chromium-review.googlesource.com/506776 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 5978e28d 2017-06-02T11:49:31 D3D11: Consolidate InputLayout allocation. This introduces a helper type to act as the initialization data type for input element arrays. A WrappedArray class acts as a helper class to wrap a C array and size into a C++ class, similar to std::array but without the data storage. Also move resource deallocation memory counting into a single code path. BUG=angleproject:2034 Change-Id: I5e108254777f0df65f5f60fe26e760c71b95b542 Reviewed-on: https://chromium-review.googlesource.com/506775 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Corentin Wallez e7557744 2017-06-01T13:09:57 WebGL compatibility: remove UB for draw buffers in the GL backend. WebGL adds two rules: - Fragment outputs declared but not written to should default to black. - FBO attachments for outputs not declared in the shader should not be written to (it is UB in OpenGL ES). Fix the first one by using the SH_INIT_OUTPUT_VARIABLES compiler options, and the second one by messing with glDrawBuffers so that the enabled draw buffers are always a subset of the ones declared by the shader. BUG=angleproject:2048 Change-Id: I1d851c190959c1acfc3e41d837e6990aec1d4086 Reviewed-on: https://chromium-review.googlesource.com/521682 Commit-Queue: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Jamie Madill b7d5e303 2017-06-01T16:04:46 D3D11: Default init all textures. The resource manager factory methods will use ClearView or zero-filled buffers to set initial data. It assumes there are no 3D depth/stencil textures. This will lead to some wasteful re-creation of RTVs in some cases. This is a temporary measure until we can implement more efficient lazy resource init strategies. BUG=angleproject:1635 Change-Id: I590e76587d3d96a359beedb79e21d24930e5f2e0 Reviewed-on: https://chromium-review.googlesource.com/503254 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill c564c070 2017-06-01T12:45:42 Pass gl::Context to impl methods instead of ContextImpl. In some cases we might have to call back into the GL layer, passing the Context, and if we just have a ContextImpl pointer this isn't possible. It also removes the need for SafeGetImpl. BUG=angleproject:2044 Change-Id: I6363e84b25648c992c25779d4c43f795aa2866d6 Reviewed-on: https://chromium-review.googlesource.com/516835 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 948bbe51 2017-06-01T13:10:42 Redesign robust resource init as a display extension. Also correct the enum allocation to values that are available to ANGLE. BUG=angleproject:1635 Change-Id: I443f5654aa6a5049b4a1ae4c253cd6473b4e446e Reviewed-on: https://chromium-review.googlesource.com/520002 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Jamie Madill cd930cf0 2017-05-31T13:36:32 EGLDevice: Fix edge-case memory leak. This could happen if there was an error on device init. Reported by Microsoft. BUG=None Change-Id: I059fd9308d0536b4120c28e684e37567d14b3fd5 Reviewed-on: https://chromium-review.googlesource.com/519642 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Jamie Madill ea19b4ac 2017-05-25T14:48:34 Don't store Renderer in RenderStateCache. Instead pass it around via methods. This makes the code a bit nicer. BUG=angleproject:2044 Change-Id: I721e190a2ecde2b1a65e57debf419ee06a5dce29 Reviewed-on: https://chromium-review.googlesource.com/516385 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Jamie Madill 9216a6e2 2017-05-24T15:53:20 Introduce MRUCache. This library comes from Chromium's base, and is useful for many use cases in ANGLE. It can replace the custom MRU code we use in the RenderStateCache. It will also be useful for implementing a program binary cache. BUG=angleproject:2044 Change-Id: Iba166fe380d7ed4e3123428b0227b9d299f756d1 Reviewed-on: https://chromium-review.googlesource.com/516384 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Jamie Madill 0a82f2fa 2017-05-30T15:42:00 D3D11: Don't copy SharedSRVs around. This was causing a very large performance regression in Texture re-binding. Instead pass by const &, which should be fine, since values in std::map (and unordered_map) are not reallocated when the map changes. Also make the SharedResource type non-copyable, and add an explicit makeCopy method for when we need to clone the shared pointer. BUG=angleproject:2034 BUG=chromium:727318 Change-Id: I39508a6ca4b41e4da31fe68899caa4464138cada Reviewed-on: https://chromium-review.googlesource.com/517670 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Jamie Madill 63d8c266 2017-05-29T13:35:09 Re-use std::hash in RenderStateCache. This will allow us to more easily take advantage of Chromium's MRUCache class. BUG=angleproject:2044 Change-Id: I3fad82fc825861dc1c2095f25da83159da76e76b Reviewed-on: https://chromium-review.googlesource.com/517359 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill aa0a5446 2017-05-25T13:30:23 Ensure gl State structs are zero filled. In some cases we would hash or memcmp against structs with bools or other non-filled data. This could have implementation differences, and may have been causing cache errors on Clang. BUG=chromium:721648 BUG=angleproject:2044 Change-Id: I981a1e6e8d50a33f7fade568497b72b919accfce Reviewed-on: https://chromium-review.googlesource.com/516383 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Geoff Lang 80616218 2017-05-17T15:40:03 Support CHROMIUM_copy_texture for all formats on D3D11. Adds a CPU readback and conversion path when the destination texture is not renderable. BUG=angleproject:1932 Change-Id: I71461ca991dc10dd636ff38e1ae20db2be0f8d63 Reviewed-on: https://chromium-review.googlesource.com/508308 Commit-Queue: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill 98de826c 2017-05-29T13:01:02 Platform: Fix initialization order in tests. Because of the platform refactor in abf38572062b91 we broke the way the tests override the D3D workarounds for the tiny depth/stencil bug. This change passes a pointer to the platform directly in the EGL init logic, which solves this issue. It also removes all decltype code in the platform header which might fix the UBSAN problem we were seeing previously. Also change a present mode selection error into a warning in the Vk back-end, since this was being triggered on AMD, but is safe. BUG=angleproject:2042 Change-Id: Ibbd0c69ce11a840cf4b33c616f56020001e553aa Reviewed-on: https://chromium-review.googlesource.com/513519 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 20805650 2017-05-25T12:20:59 D3D11: Consolidate state allocation. This cleans up the allocation and deallocation of Blend, DepthStencil, Rasterizer, and Sampler states. This patch introduces a LazyResource2 class, basically a replacement for LazyResource, which will be removed once the refactor is done. BUG=angleproject:2034 Change-Id: I4fa759ae479807ff69a629f89a08b01800ba3f66 Reviewed-on: https://chromium-review.googlesource.com/503627 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 9733ceef 2017-05-11T19:14:35 Initialize uninitialized locals in GLSL output Guarantee that local variables are initialized before they are used in GLSL output. In HLSL output all variables were already being initialized. Locals are initialized using an AST transform. The local variable init can only be run after some simplification of the AST, so that it is able to handle complex cases like: for (int i[2], j = i[0]; i[0] < 3; ++i[0]) { } If we're dealing with ESSL 1.00 which lacks array constructors, in this kind of case the uninitialized array initialization code needs to be hoisted out of the loop init statement, and the code also needs to make sure that j's initializer is run after i is initialized. Another complex case involves nameless structs. This can be an issue also in ESSL 3.00 and above: for (struct { float f; } s; s.f < 1.0; ++s.f) { } Since the struct doesn't have a name, its constructor can not be used. We solve this by initializing the struct members individually, similarly to how arrays are initialized in ESSL 1.00. Initializing local variables is disabled on Mac and Android for now. On Mac, invalid behavior was exposed in the WebGL 2.0 tests when enabling it. On Android, the dEQP test runs failed for an unknown reason. Bugs have been opened to resolve these issues later. BUG=angleproject:1966 TEST=angle_end2end_tests, WebGL conformance tests Change-Id: Ic06927f5b6cc9619bc82c647ee966605cd80bab2 Reviewed-on: https://chromium-review.googlesource.com/504728 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill bdc1e2a2 2017-05-24T14:40:27 D3D11: Consolidate Buffer allocation. We now allocate and release all buffer resources from a single place, allowing us to allocate junk memory, clear on init, and track the currently allocated buffer memory. BUG=angleproject:2034 Change-Id: Id60b63a7f77bc369dfc494a7587ab06c8d34a8e2 Reviewed-on: https://chromium-review.googlesource.com/503253 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 2c479d6e 2017-05-24T14:37:22 D3D11: Use TextureHelper11 everywhere. This consolidates all texture allocation into ResourceManager11. It removes a lot of error checking and resource management code. In a few places we're storing some redundant information, like in TextureStorage11, we might store the Format pointer in two places, or the Texture's size. BUG=angleproject:2034 Change-Id: I9369e76925a67632c444c662e5667c5ed7875547 Reviewed-on: https://chromium-review.googlesource.com/503252 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 6811a316 2017-05-24T16:50:06 Use GetPlatformDisplay instead of GetPlatformDisplayEXT. This will allow us to use EGLAttrib and pass pointers as Display init parameters if necessary. BUG=angleproject:2042 Change-Id: Ib0f85d71bc9d3dff2db9453012f40303f3c3bd7c Reviewed-on: https://chromium-review.googlesource.com/513518 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 9d7d719c 2017-05-24T14:35:59 D3D11: Move CreateStagingTexture to Renderer11. This will make staging texture tracking easier. BUG=angleproject:2034 Change-Id: I2f0418cc5784100bf6fba59b4bda1b2dd9dc18d1 Reviewed-on: https://chromium-review.googlesource.com/503251 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Geoff Lang 4f0e003e 2017-05-01T16:04:35 Implement the new formats/features of the ES3 CHROMIUM_copy_texture. Some non-renderable texture formats remain unimplemented. BUG=angleproject:1932 Change-Id: Id206432d6e26a70fc0e84478a4e43e9eefadcf2f Reviewed-on: https://chromium-review.googlesource.com/491948 Commit-Queue: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill caafd890 2017-05-19T14:31:25 D3D11: Don't store device in InputLayoutCache. We will need the Renderer pointer for resource allocation, so pass it around instead of storing the device and context. BUG=angleproject:2034 Change-Id: Iaf271913f88c5402895aa81de310aa22fcd72cc4 Reviewed-on: https://chromium-review.googlesource.com/509930 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill e0e009fe 2017-05-19T14:13:03 D3D11: Consolidate SRV allocation. This change also moves all SRV storage to use the custom smart pointer classes in ResourceManager11. Every single SRV is now allocated in ResourceManager11. For TextureStorage11 and RenderTarget11, we use a shared pointer type since the SRVs use shared ownership by both classes. BUG=angleproject:2034 Change-Id: I90666fbbbc23ff10870b7401f3da09ef79258136 Reviewed-on: https://chromium-review.googlesource.com/503250 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 39b826fc 2017-05-24T09:47:44 Reduce warning spam for failed D3D shaders. This prevents Chrome's about:gpu from being filled with bad shaders, and instead only prints the relevant error lines. BUG=None Change-Id: I75fcd2ef920d2ed7b98e8668c487b0921faf4c0b Reviewed-on: https://chromium-review.googlesource.com/513417 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 764a1e76 2017-05-18T11:11:47 D3D11: Consolidate DSV allocation. This uses the new classes introduced in ResourceManager11. Smart pointers mean we can use ANGLE_TRY in more places, and don't need to manually release these views. There is now a single point in the D3D11 back-end where depth stencil views are allocated. BUG=angleproject:2034 Change-Id: If0de2529ece1298e8babf24c07cc4be9849cc4de Reviewed-on: https://chromium-review.googlesource.com/503249 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Shao df682a8e 2017-03-31T15:13:21 ES31: Refactor syncState in VertexArrayGL This patch intends to solve several design issues by refactoring the process of syncState in VertexArrayGL before implementing ES3.1 feature Vertex Attrib Binding on the OpenGL back-end. 1. Use nullptr as the flag of using client memory pointer 2. Simplify comparisons in updateAttribPointer. 3. Put all code related to mFunctions->vertexAttrib*Pointer() into an individual function 4. Remove redundant mStateManager->bindVertexArray() in all update* functioins and only call it once in syncState(). BUG=angleproject:1593 Change-Id: I8f68534bb9291a601b9b77954d7281e5171c2b55 Reviewed-on: https://chromium-review.googlesource.com/465378 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill 403c1682 2017-05-18T11:11:46 D3D11: Add resource manager classes. The resource manager tracks the allocations and deallocations of all D3D11 resources that have device memory: Buffers and Textures (which are either 2D or 3D). It also tracks the number of active Views (DSV, RTV, and SRV, potentially UAV with ES 3.1). A new smart pointer type will wrap the resource deallocation so that the object notifies the manager when ANGLE is done with the resouce. This allows us to track precisely how much GPU memory we think we're using at any point, and will help prevent resource leaks for these object types. It also makes initialization and releasing much more trivial. The base class for a resource uses a template template parameter so that we can use a unique or shared pointer depending on if the object in question needs unique or shared ownership. For some resources (in our case, SRVs are shared between the TextureStorage11 and RenderTarget11 classes, and Textures are shared in many places) we need to have the ability to have shared ownership. Unique ownership is a little bit more efficient so supporting both can be helpful. In this patch RenderTargetView allocation is moved to use the unique smart pointer. BUG=angleproject:2034 Change-Id: Idb1245c24cd66733b8b5ca524c727350b2d2c745 Reviewed-on: https://chromium-review.googlesource.com/503248 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Geoff Lang 451b2b7f 2017-05-15T15:32:02 Correctly generate mipmaps for sRGB textures. sRGB textures need to linearize their data before averaging, this worked correctly on the GPU mipmap generation functions but not the CPU ones. BUG=angleproject:2019 TEST=conformance/extensions/ext-sRGB Change-Id: I554dac89b12acf5ebf5b7cab6f6faf0bce5168a6 Reviewed-on: https://chromium-review.googlesource.com/506241 Commit-Queue: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill 231c7f56 2017-04-26T13:45:37 Apply clang-format to many files. This cleans up the formatting in many places. BUG=None Change-Id: I6c6652ebc042f1f0ffecced53582d09d66b4f384 Reviewed-on: https://chromium-review.googlesource.com/487884 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Xinghua Cao 73badc07 2017-03-29T19:14:53 ES31: Implement glDispatchCompute for D3D backend BUG=angleproject:1955 TESTCASE=angle_end2end_tests Change-Id: I69b4df83d67017d39df67753d6d17fc15ececebf Reviewed-on: https://chromium-review.googlesource.com/462067 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jiawei-Shao 995c2ed2 2017-05-15T13:31:45 Initialize currentValueAttrib->binding in updateCurrentValueAttribs This patch intends to fix a crash issue in the win-clang build of ANGLE by adding the missing currentValueAttrib->binding assignment in StateManager11.cpp::updateCurrentValueAttribs(). In D3D11 all vertex attributes (VertexAttribute and its VertexBinding) should be treated as a whole, so we should always ensure these two parts are availabe. This patch also add ASSERTs before using *translated->attrib and *translated->binding to prevent a TranslatedAttribute object having a null attrib or binding. BUG=chromium:721783, angleproject:1593 Change-Id: Id11d6f1d4c37daabad2265e01ea38eb51046b9cb Reviewed-on: https://chromium-review.googlesource.com/505928 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Corentin Wallez <cwallez@chromium.org>
kmx.io     mail     discord kmxgit v0.4.0